Module:User:JeromeWillettID/sandbox

local export = {}

--[=[

Authorship: Ben Wing

]=]

--[=[

TERMINOLOGY:

-- "slot" = A particular combination of tense/mood/person/number/etc. Example slot names for verbs are "pres_1s" (present indicative first-person singular), "pres_sub_2s" (present	 subjunctive second-person singular) "impf_sub_3p" (imperfect subjunctive third-person plural). Each slot is filled with zero or more forms.

-- "form" = The conjugated Portuguese form representing the value of a given slot.

-- "lemma" = The dictionary form of a given Portuguese term. For Portuguese, always the infinitive. ]=]

--[=[

FIXME:

--]=]

local lang = require("Module:languages").getByCode("pt") local m_string_utilities = require("Module:string utilities") local m_links = require("Module:links") local m_table = require("Module:table") local iut = require("Module:inflection utilities") local com = require("Module:pt-common")

local force_cat = false -- set to true for debugging local check_for_red_links = false -- set to false for debugging

local rfind = mw.ustring.find local rmatch = mw.ustring.match local rsplit = mw.text.split local rsub = com.rsub local u = mw.ustring.char

local function link_term(term) return m_links.full_link({ lang = lang, term = term }, "term") end

local V = com.V -- vowel regex class local AV = com.AV -- accented vowel regex class local C = com.C -- consonant regex class

local AC = u(0x0301) -- acute = ́ local TEMPC1 = u(0xFFF1) -- temporary character used for consonant substitutions local TEMP_MESOCLITIC_INSERTION_POINT = u(0xFFF2) -- temporary character used to mark the mesoclitic insertion point local VAR_BR = u(0xFFF3) -- variant code for Brazil local VAR_PT = u(0xFFF4) -- variant code for Portugal local VAR_SUPERSEDED = u(0xFFF5) -- variant code for superseded forms local VAR_NORMAL = u(0xFFF6) -- variant code for non-superseded forms local all_var_codes = VAR_BR .. VAR_PT .. VAR_SUPERSEDED .. VAR_NORMAL local var_codes_no_superseded = VAR_BR .. VAR_PT .. VAR_NORMAL local var_code_c = "[" .. all_var_codes .. "]" local var_code_no_superseded_c = "[" .. var_codes_no_superseded .. "]" local not_var_code_c = "[^" .. all_var_codes .. "]"

-- Export variant codes for use in Module:pt-inflections. export.VAR_BR = VAR_BR export.VAR_PT = VAR_PT export.VAR_SUPERSEDED = VAR_SUPERSEDED export.VAR_NORMAL = VAR_NORMAL

local short_pp_footnote = "[usually used with auxiliary verbs " .. link_term("ser") .. " and " .. link_term("estar") .. "]" local long_pp_footnote = "[usually used with auxiliary verbs " .. link_term("haver") .. " and " .. link_term("ter") .. "]"

--[=[

Vowel alternations:

: 'i' in pres1s and the whole present subjunctive; 'e' elsewhere when stressed. Generally 'e' otherwise when unstressed. E.g. sentir, conseguir (the latter additionally with 'gu-g' alternation). : 'u' in pres1s and the whole present subjunctive; 'o' elsewhere when stressed. Either 'o' or 'u' otherwise when unstressed. E.g. dormir, subir. : 'i' whenever stressed (in the present singular and third plural) and throughout the whole present subjunctive. Otherwise 'e'. E.g. progredir, also premir per Priberam. : 'u' whenever stressed (in the present singular and third plural) and throughout the whole present subjunctive. Otherwise 'o'. E.g. polir, extorquir (the latter also ). <í>: The last 'i' of the stem (excluding stem-final 'i') becomes 'í' when stressed. E.g.: * proibir ('proíbo, proíbe(s), proíbem, proíba(s), proíbam') * faiscar ('faísco, faísca(s), faíscam, faísque(s), faísquem' also with 'c-qu' alternation) * homogeneizar ('homogeneízo', etc.) * mobiliar ('mobílio', etc.; note here the final -i is ignored when determining which vowel to stress) * tuitar ('tuíto', etc.) <ú>: The last 'u' of the stem (excluding stem-final 'u') becomes 'ú' when stressed. E.g.: * reunir ('reúno, reúne(s), reúnem, reúna(s), reúnam') * esmiuçar ('esmiúço, esmiúça(s), esmiúça, esmiúce(s), esmiúcem' also with 'ç-c' alternation) * reusar ('reúso, reúsa(s), reúsa, reúse(s), reúsem') * saudar ('saúdo, saúda(s), saúda, saúde(s), saúdem') ]=]

local vowel_alternants = m_table.listToSet({"i-e", "i", "í", "u-o", "u", "ú", "ei", "+"}) local vowel_alternant_to_desc = { ["i-e"] = "i-e alternation in present singular", ["i"] = "e becomes i when stressed", ["í"] = "i becomes í when stressed", ["u-o"] = "u-o alternation in present singular", ["u"] = "o becomes u when stressed", ["ú"] = "u becomes ú when stressed", ["ei"] = "i becomes ei when stressed", }

local vowel_alternant_to_cat = { ["i-e"] = "i-e alternation in present singular", ["i"] = "e becoming i when stressed", ["í"] = "i becoming í when stressed", ["u-o"] = "u-o alternation in present singular", ["u"] = "o becoming u when stressed", ["ú"] = "u becoming ú when stressed", ["ei"] = "i becoming ei when stressed", }

local all_persons_numbers = { ["1s"] = "1|s", ["2s"] = "2|s", ["3s"] = "3|s", ["1p"] = "1|p", ["2p"] = "2|p", ["3p"] = "3|p", }

local person_number_list = {"1s", "2s", "3s", "1p", "2p", "3p"} local imp_person_number_list = {"2s", "3s", "1p", "2p", "3p"} local neg_imp_person_number_list = {"2s", "3s", "1p", "2p", "3p"}

person_number_to_reflexive_pronoun = { ["1s"] = "me", ["2s"] = "te", ["3s"] = "se", ["1p"] = "nos", ["2p"] = "vos", ["3p"] = "se", }

local indicator_flags = m_table.listToSet { "no_pres_stressed", "no_pres1_and_sub", "only3s", "only3sp", "only3p", "pp_inv", "irreg", "no_built_in", "e_ei_cat", }

-- Remove any variant codes e.g. VAR_BR, VAR_PT, VAR_SUPERSEDED. Needs to be called from Module:pt-headword on the -- output of do_generate_forms. `keep_superseded` leaves VAR_SUPERSEDED; used in the `canonicalize` function of -- show_forms because we then process and remove it in `generate_forms`. FIXME: Use metadata for this once it's -- supported in Module:inflection utilities. function export.remove_variant_codes(form, keep_superseded) return rsub(form, keep_superseded and var_code_no_superseded_c or var_code_c, "") end

-- Initialize all the slots for which we generate forms. local function add_slots(alternant_multiword_spec) -- "Basic" slots: All slots that go into the regular table (not the reflexive form-of table). alternant_multiword_spec.verb_slots_basic = { {"infinitive", "inf"}, {"infinitive_linked", "inf"}, {"gerund", "ger"}, {"short_pp_ms", "short|m|s|past|part"}, {"short_pp_fs", "short|f|s|past|part"}, {"short_pp_mp", "short|m|p|past|part"}, {"short_pp_fp", "short|f|p|past|part"}, {"pp_ms", "m|s|past|part"}, {"pp_fs", "f|s|past|part"}, {"pp_mp", "m|p|past|part"}, {"pp_fp", "f|p|past|part"}, }

-- Special slots used to handle non-reflexive parts of reflexive verbs in. -- For example, for a reflexive-only verb like esbaldar-se, we want to be able to use on	-- esbalde (which should mention that it is a part of 'me esbalde', first-person singular present subjunctive,	-- and 'se esbalde', third-person singular present subjunctive) or on esbaldamos (which should mention that it	-- is a part of 'esbaldamo-nos', first-person plural present indicative or preterite). Similarly, we want to use -- on esbaldando (which should mention that it is a part of 'se ... esbaldando', syntactic	-- variant of esbaldando-se, which is the gerund of esbaldar-se). To do this, we need to be able to map -- non-reflexive parts like esbalde, esbaldamos, esbaldando, etc. to their reflexive equivalent(s), to	-- the tag(s) of the equivalent(s), and, in the case of forms like esbaldando, esbaldar and imperatives, to -- the separated syntactic variant of the verb+clitic combination. We do this by creating slots for the -- non-reflexive part equivalent of each basic reflexive slot, and for the separated syntactic-variant equivalent -- of each basic reflexive slot that is formed of verb+clitic. We use slots in this way to deal with multiword -- lemmas. Note that we run into difficulties mapping between reflexive verbs, non-reflexive part equivalents, and -- separated syntactic variants if a slot contains more than one form. To handle this, if there are the same number -- of forms in two slots we're trying to match up, we assume the forms match one-to-one; otherwise we don't match up	-- the two slots (which means won't work in this case, but such a case is extremely rare and not	-- worth worrying about). Alternatives that handle this "properly" are significantly more complicated and require -- non-trivial modifications to Module:inflection utilities. local need_special_verb_form_of_slots = alternant_multiword_spec.from_verb_form_of and alternant_multiword_spec.refl

if need_special_verb_form_of_slots then alternant_multiword_spec.verb_slots_reflexive_verb_form_of = { {"infinitive_non_reflexive", "-"}, {"infinitive_variant", "-"}, {"gerund_non_reflexive", "-"}, {"gerund_variant", "-"}, }	else alternant_multiword_spec.verb_slots_reflexive_verb_form_of = {} end

-- Add entries for a slot with person/number variants. -- `verb_slots` is the table to add to. -- `slot_prefix` is the prefix of the slot, typically specifying the tense/aspect. -- `tag_suffix` is a string listing the set of inflection tags to add after the person/number tags. -- `person_number_list` is a list of the person/number slot suffixes to add to `slot_prefix`. local function add_personal_slot(verb_slots, slot_prefix, tag_suffix, person_number_list) for _, persnum in ipairs(person_number_list) do			local persnum_tag = all_persons_numbers[persnum] local slot = slot_prefix .. "_" .. persnum local accel = persnum_tag .. "|" .. tag_suffix table.insert(verb_slots, {slot, accel}) end end

-- Add a personal slot (i.e. a slot with person/number variants) to `verb_slots_basic`. local function add_basic_personal_slot(slot_prefix, tag_suffix, person_number_list, no_special_verb_form_of_slot) add_personal_slot(alternant_multiword_spec.verb_slots_basic, slot_prefix, tag_suffix, person_number_list) -- Add special slots for handling non-reflexive parts of reflexive verbs in. -- See comment above in `need_special_verb_form_of_slots`. if need_special_verb_form_of_slots and not no_special_verb_form_of_slot then for _, persnum in ipairs(person_number_list) do				local persnum_tag = all_persons_numbers[persnum] local basic_slot = slot_prefix .. "_" .. persnum local accel = persnum_tag .. "|" .. tag_suffix table.insert(alternant_multiword_spec.verb_slots_reflexive_verb_form_of, {basic_slot .. "_non_reflexive", "-"}) end end end

add_basic_personal_slot("pres", "pres|ind", person_number_list) add_basic_personal_slot("impf", "impf|ind", person_number_list) add_basic_personal_slot("pret", "pret|ind", person_number_list) add_basic_personal_slot("plup", "plup|ind", person_number_list) add_basic_personal_slot("fut", "fut|ind", person_number_list) add_basic_personal_slot("cond", "cond", person_number_list) add_basic_personal_slot("pres_sub", "pres|sub", person_number_list) add_basic_personal_slot("impf_sub", "impf|sub", person_number_list) add_basic_personal_slot("fut_sub", "fut|sub", person_number_list) add_basic_personal_slot("imp", "imp", imp_person_number_list) add_basic_personal_slot("pers_inf", "pers|inf", person_number_list) -- Don't need special non-reflexive-part slots because the negative imperative is multiword, of which the -- individual words are 'não' + subjunctive. add_basic_personal_slot("neg_imp", "neg|imp", neg_imp_person_number_list, "no special verb form of") -- Don't need special non-reflexive-part slots because we don't want esbaldando mapping to esbaldando-me -- (only esbaldando-se) or esbaldar mapping to esbaldar-me (only esbaldar-se). add_basic_personal_slot("infinitive", "inf", person_number_list, "no special verb form of") add_basic_personal_slot("gerund", "ger", person_number_list, "no special verb form of")

-- Generate the list of all slots. alternant_multiword_spec.all_verb_slots = {} for _, slot_and_accel in ipairs(alternant_multiword_spec.verb_slots_basic) do		table.insert(alternant_multiword_spec.all_verb_slots, slot_and_accel) end for _, slot_and_accel in ipairs(alternant_multiword_spec.verb_slots_reflexive_verb_form_of) do		table.insert(alternant_multiword_spec.all_verb_slots, slot_and_accel) end

alternant_multiword_spec.verb_slots_basic_map = {} for _, slotaccel in ipairs(alternant_multiword_spec.verb_slots_basic) do		local slot, accel = unpack(slotaccel) alternant_multiword_spec.verb_slots_basic_map[slot] = accel end end

local overridable_stems = {}

local function allow_multiple_values(separated_groups, data) local retvals = {} for _, separated_group in ipairs(separated_groups) do		local footnotes = data.fetch_footnotes(separated_group) local retval = {form = separated_group[1], footnotes = footnotes} table.insert(retvals, retval) end return retvals end

local function simple_choice(choices) return function(separated_groups, data) if #separated_groups > 1 then data.parse_err("For spec '" .. data.prefix .. ":', only one value currently allowed") end if #separated_groups[1] > 1 then data.parse_err("For spec '" .. data.prefix .. ":', no footnotes currently allowed") end local choice = separated_groups[1][1] if not m_table.contains(choices, choice) then data.parse_err("For spec '" .. data.prefix .. ":', saw value '" .. choice .. "' but expected one of '" ..				table.concat(choices, ",") .. "'") end return choice end end

for _, overridable_stem in ipairs { "pres_unstressed", "pres_stressed", "pres1_and_sub", -- Don't include pres1; use pres_1s if you need to override just that form "impf", "full_impf", "pret_base", "pret", {"pret_conj", simple_choice({"irreg", "ar", "er", "ir"}) }, "fut", "cond", "pres_sub_stressed", "pres_sub_unstressed", {"sub_conj", simple_choice({"ar", "er"}) }, "plup", "impf_sub", "fut_sub", "pers_inf", "pp", "short_pp", } do	if type(overridable_stem) == "string" then overridable_stems[overridable_stem] = allow_multiple_values else local stem, validator = unpack(overridable_stem) overridable_stems[stem] = validator end end

-- Useful as the value of the `match` property of a built-in verb. `main_verb_spec` is a Lua pattern that should match -- the non-prefixed part of a verb, and `prefix_specs` is a list of Lua patterns that should match the prefixed part of -- a verb. If a prefix spec is preceded by ^, it must match exactly at the beginning of the verb; otherwise, additional -- prefixes (e.g. re-, des-) may precede. Return the prefix and main verb. local function match_against_verbs(main_verb_spec, prefix_specs) return function(verb) for _, prefix_spec in ipairs(prefix_specs) do			if prefix_spec:find("^%^") then -- must match exactly prefix_spec = prefix_spec:gsub("^%^", "") if prefix_spec == "" then -- We can't use the second branch of the if-else statement because an empty returns the current position -- in rmatch. local main_verb = rmatch(verb, "^(" .. main_verb_spec .. ")$")					if main_verb then return "", main_verb end else local prefix, main_verb = rmatch(verb, "^(" .. prefix_spec .. ")(" .. main_verb_spec .. ")$")					if prefix then return prefix, main_verb end end else local prefix, main_verb = rmatch(verb, "^(.*" .. prefix_spec .. ")(" .. main_verb_spec .. ")$")				if prefix then return prefix, main_verb end end end return nil end end

--[=[

Built-in (usually irregular) conjugations.

Each entry is processed in turn and consists of an object with two fields: 1. match=: Specifies the built-in verbs that match this object. 2. forms=: Specifies the built-in stems and forms for these verbs.

The value of match= is either a string beginning with "^" (match only the specified verb), a string not beginning with "^" (match any verb ending in that string), or a function that is passed in the verb and should return the prefix of the verb if it matches, otherwise nil. The function match_against_verbs is provided to facilitate matching a set of verbs with a common ending and specific prefixes (e.g. ter and ater but not abater, etc.).

The value of forms= is a table specifying stems and individual override forms. Each key of the table names either a stem (e.g. `pres_stressed`), a stem property (e.g. `vowel_alt`) or an individual override form (e.g. `pres_1s`). Each value of a stem can either be a string (a single stem), a list of strings, or a list of objects of the form {form = STEM, footnotes = {FOONOTES}}. Each value of an individual override should be of exactly the same form except that the strings specify full forms rather than stems. The values of a stem property depend on the specific property but are generally strings or booleans.

In order to understand how the stem specifications work, it's important to understand the phonetic modifications done by combine_stem_ending. In general, the complexities of predictable prefix, stem and ending modifications are all handled in this function. In particular:

1. Spelling-based modifications (c/z, g/gu, gu/gü, g/j) occur automatically as appropriate for the ending. 2. If the stem begins with an acute accent, the accent is moved onto the last vowel of the prefix (for handling verbs  in -uar such as minguar, pres_3s 'míngua'). 3. If the ending begins with a double asterisk, this is a signal to conditionally delete the accent on the last letter of the stem. "Conditionally" means we don't do it if the last two letters would form a diphthong without the accent on the second one (e.g. in sair, with stem 'saí'); but as an exception, we do delete the accent in stems ending in -guí, -quí (e.g. in conseguir) because in this case the ui isn't a diphthong. 4. If the ending begins with an asterisk, this is a signal to delete the accent on the last letter of the stem, e.g.  fizé -> fizermos. Unlike for **, this removal is unconditional, so we get e.g. 'sairmos' not #'saírmos'. 5. If ending begins with i, it must get an accent after an unstressed vowel (in some but not all cases) to prevent the two merging into a diphthong. See combine_stem_ending for specifics.

The following stems are recognized:

-- pres_unstressed: The present indicative unstressed stem (1p, 2p). Also controls the imperative 2p and gerund. Defaults to the infinitive stem (minus the ending -ar/-er/-ir/-or). -- pres_stressed: The present indicative stressed stem (1s, 2s, 3s, 3p). Also controls the imperative 2s. Default is empty if indicator `no_pres_stressed`, else a vowel alternation if such an indicator is given (e.g. `ue`, `ì`), else the infinitive stem. -- pres1_and_sub: Overriding stem for 1s present indicative and the entire subjunctive. Only set by irregular verbs and by the indicators `no_pres_stressed` (e.g. precaver) and `no_pres1_and_sub` (since verbs of this sort,	 e.g. puir, are missing the entire subjunctive as well as the 1s present indicative). Used by many irregular verbs, e.g. caber, verbs in '-air', dizer, ter, valer, etc. Some verbs set this and then supply an override for the pres_1sg if it's irregular, e.g. saber, with irregular subjunctive stem "saib-" and special 1s present indicative "sei". -- pres1: Special stem for 1s present indicative. Normally, do not set this explicitly. If you need to specify an    irregular 1s present indicative, use the form override pres_1s= to specify the entire form. Defaults to pres1_and_sub if given, else pres_stressed. -- pres_sub_unstressed: The present subjunctive unstressed stem (1p, 2p). Defaults to pres1_and_sub if given, else the infinitive stem. -- pres_sub_stressed: The present subjunctive stressed stem (1s, 2s, 3s, 1p). Defaults to pres1. -- sub_conj: Determines the set of endings used in the subjunctive. Should be one of "ar" or "er". -- impf: The imperfect stem (not including the -av-/-i- stem suffix, which is determined by the conjugation). Defaults to the infinitive stem. -- full_impf: The full imperfect stem missing only the endings (-a, -as, -am, etc.). Used for verbs with irregular imperfects such as ser, ter, vir and pôr. Overrides must be supplied for the impf_1p and impf_2p due to these forms having an accent on the stem. -- pret_base: The preterite stem (not including the -a-/-e-/-i- stem suffix). Defaults to the infinitive stem. -- pret: The full preterite stem missing only the endings (-ste, -mos, etc.). Used for verbs with irregular preterites (pret_conj == "irreg") such as fazer, poder, trazer, etc. Overrides must be supplied for the pret_1s and pret_3s. Defaults to `pret_base` + the accented conjugation vowel. -- pret_conj: Determines the set of endings used in the preterite. Should be one of "ar", "er", "ir" or "irreg". Defaults to the conjugation as determined from the infinitive. When pret_conj == "irreg", stem `pret` is used, otherwise `pret_base`. -- fut: The future stem. Defaults to the infinitive stem + the unaccented conjugation vowel. -- cond: The conditional stem. Defaults to `fut`. -- impf_sub: The imperfect subjunctive stem. Defaults to `pret`. -- fut_sub: The future subjunctive stem. Defaults to `pret`. -- plup: The pluperfect stem. Defaults to `pret`. -- pers_inf: The personal infinitive stem. Defaults to the infinitive stem + the accented conjugation vowel. -- pp: The masculine singular past participle. Default is based on the verb conjugation: infinitive stem + "ado" for -ar verbs, otherwise infinitive stem + "ido". -- short_pp: The short masculine singular past participle, for verbs with such a form. No default. -- pp_inv: True if the past participle exists only in the masculine singular. ]=]

local built_in_conjugations = {

--                                            -ar                                        --

-- Verbs not needing entries here: --	-- (1) Verbs with short past participles: need to specify the short pp explicitly. --	-- aceitar: use  -- anexar, completar, expressar, expulsar, findar, fritar, ganhar, gastar, limpar, pagar, pasmar, pegar, soltar: --  use  etc.	-- assentar: use  -- entregar: use  -- enxugar: use  -- matar: use  --	-- (2) Verbs with orthographic consonant alternations: handled automatically. --	-- -car (brincar, buscar, pecar, trancar, etc.): automatically handled in combine_stem_ending -- -çar (alcançar, começar, laçar): automatically handled in combine_stem_ending -- -gar (apagar, cegar, esmagar, largar, navegar, resmungar, sugar, etc.): automatically handled in combine_stem_ending --	-- (3) Verbs with vowel alternations: need to specify the alternation explicitly unless it always happens, in --    which case it's handled automatically through an entry below. --	-- esmiuçar changing to esmiúço: use <ú> -- faiscar changing to faísco: use <í> -- -iar changing to -eio (ansiar, incendiar, mediar, odiar, remediar, etc.): use  -- -izar changing to -ízo (ajuizar, enraizar, homogeneizar, plebeizar, etc.): use <í> -- mobiliar changing to mobílio: use <í> -- reusar changing to reúso: use <ú> -- saudar changing to saúdo: use <ú> -- tuitar/retuitar changing to (re)tuíto: use <í>

{		-- dar, desdar match = match_against_verbs("dar", {"^", "des"}), forms = { pres_1s = "dou", pres_2s = "dás", pres_3s = "dá", -- damos, dais regular pres_3p = "dão", pret = "dé", pret_conj = "irreg", pret_1s = "dei", pret_3s = "deu", pres_sub_1s = "dê", pres_sub_2s = "dês", pres_sub_3s = "dê", pres_sub_1p = {"demos", VAR_SUPERSEDED .. "dêmos"}, -- deis regular pres_sub_3p = {"deem", VAR_SUPERSEDED .. "dêem"}, irreg = true, }	},	{		-- -ear (frear, nomear, semear, etc.) match = "ear", forms = { pres_stressed = "ei", e_ei_cat = true, }	},	{		-- estar match = match_against_verbs("estar", {"^", "sob", "sobr"}), forms = { pres_1s = "estou", pres_2s = "estás", pres_3s = "está", -- FIXME, estámos is claimed as an alternative pres_1p in the old conjugation data, but I believe this is garbage pres_3p = "estão", pres1_and_sub = "estej", -- only for subjunctive as we override pres_1s sub_conj = "er", pret = "estivé", pret_conj = "irreg", pret_1s = "estive", pret_3s = "esteve", -- sobestar, sobrestar are transitive so they have fully inflected past participles pp_inv = function(base, prefix) return prefix == "" end, irreg = true, }	},	{		-- It appears that only resfolegar has proparoxytone forms, not folegar or tresfolegar. match = "^resfolegar", forms = { pres_stressed = {"resfóleg", "resfoleg"}, irreg = true, }	},	{		-- aguar/desaguar/enxaguar, ambiguar/apaziguar/averiguar, minguar, cheguar?? (obsolete variant of chegar) match = "guar", forms = { -- combine_stem_ending will move the acute accent backwards so it sits after the last vowel in minguar pres_stressed = {{form = AC .. "gu", footnotes = {"[Brazil]"}}, {form = "gu", footnotes = {"[Portugal]"}}}, pres_sub_stressed = { {form = AC .. "gu", footnotes = {"[Brazil]"}}, {form = "gu", footnotes = {"[Portugal]"}}, {form = AC .. VAR_SUPERSEDED .. "gü", footnotes = {"[Brazil]"}}, {form = VAR_SUPERSEDED .. "gú", footnotes = {"[Portugal]"}}, },			pres_sub_unstressed = {"gu", {form = VAR_SUPERSEDED .. "gü", footnotes = {"[Brazil]"}}}, pret_1s = {"guei", {form = VAR_SUPERSEDED .. "güei", footnotes = {"[Brazil]"}}}, }	},	{		-- adequar/readequar, antiquar/obliquar, apropinquar match = "quar", forms = { -- combine_stem_ending will move the acute accent backwards so it sits after the last vowel in apropinquar pres_stressed = {{form = AC .. "qu", footnotes = {"[Brazil]"}}, {form = "qu", footnotes = {"[Portugal]"}}}, pres_sub_stressed = { {form = AC .. "qu", footnotes = {"[Brazil]"}}, {form = "qu", footnotes = {"[Portugal]"}}, {form = AC .. VAR_SUPERSEDED .. "qü", footnotes = {"[Brazil]"}}, {form = VAR_SUPERSEDED .. "qú", footnotes = {"[Portugal]"}}, },			pres_sub_unstressed = {"qu", {form = VAR_SUPERSEDED .. "qü", footnotes = {"[Brazil]"}}}, pret_1s = {"quei", {form = VAR_SUPERSEDED .. "qüei", footnotes = {"[Brazil]"}}}, }	},	{		-- -oar (abençoar, coroar, enjoar, perdoar, etc.) match = "oar", forms = { pres_1s = {"oo", VAR_SUPERSEDED .. "ôo"}, }	},	{		-- -oiar (apoiar, boiar) match = "oiar", forms = { pres_stressed = {"oi", {form = VAR_SUPERSEDED .. "ói", footnotes = {"[Brazil]"}}}, }	},	{		-- parar match = "^parar", forms = { pres_3s = {"para", VAR_SUPERSEDED .. "pára"}, }	},	{		-- pelar match = "^pelar", forms = { pres_1s = {"pelo", VAR_SUPERSEDED .. "pélo"}, pres_2s = {"pelas", VAR_SUPERSEDED .. "pélas"}, pres_3s = {"pela", VAR_SUPERSEDED .. "péla"}, }	},

--                                            -er                                        --

-- Verbs not needing entries here: --	-- precaver: use  -- -cer (verbs in -ecer, descer, vencer, etc.): automatically handled in combine_stem_ending -- -ger (proteger, reger, etc.): automatically handled in combine_stem_ending -- -guer (erguer/reerguer/soerguer): automatically handled in combine_stem_ending

{		-- benzer match = "benzer", forms = {short_pp = "bento"} },	{		-- caber match = "caber", forms = { pres1_and_sub = "caib", pret = "coubé", pret_1s = "coube", pret_3s = "coube", pret_conj = "irreg", irreg = true, }	},	{		-- crer, descrer match = "crer", forms = { pres_2s = "crês", pres_3s = "crê", pres_2p = "credes", pres_3p = {"creem", VAR_SUPERSEDED .. "crêem"}, pres1_and_sub = "crei", irreg = true, }	},	{		-- dizer, bendizer, condizer, contradizer, desdizer, maldizer, predizer, etc.		match = "dizer", forms = { -- use 'digu' because we're in a front environment; if we use 'dig', we'll get '#dijo' pres1_and_sub = "digu", pres_3s = "diz", pret = "dissé", pret_conj = "irreg", pret_1s = "disse", pret_3s = "disse", pp = "dito", fut = "dir", imp_2s = {"diz", "dize"}, -- per Infopédia irreg = true, }	},	{		-- eleger, reeleger match = "eleger", forms = {short_pp = "eleito"} },	{		-- acender, prender; not desprender, etc.		match = match_against_verbs("ender", {"^ac", "^pr"}), forms = {short_pp = "eso"} },	{		-- fazer, afazer, contrafazer, desfazer, liquefazer, perfazer, putrefazer, rarefazer, refazer, satisfazer, tumefazer match = "fazer", forms = { pres1_and_sub = "faç", pres_3s = "faz", pret = "fizé", pret_conj = "irreg", pret_1s = "fiz", pret_3s = "fez", pp = "feito", fut = "far", imp_2s = {"faz", {form = "faze", footnotes = {"[Brazil only]"}}}, -- per Priberam irreg = true, }	},	{		match = "^haver", forms = { pres_1s = "hei", pres_2s = "hás", pres_3s = "há", pres_1p = {"havemos", "hemos"}, pres_2p = {"haveis", "heis"}, pres_3p = "hão", pres1_and_sub = "haj", -- only for subjunctive as we override pres_1s pret = "houvé", pret_conj = "irreg", pret_1s = "houve", pret_3s = "houve", imp_2p = "havei", irreg = true, }	},	-- reaver below under r-	{ -- jazer, adjazer match = "jazer", forms = { pres_3s = "jaz", imp_2s = {"jaz", "jaze"}, -- per Infopédia irreg = true, }	},	{		-- ler, reler, tresler; not excel(l)er, valer, etc.		match = match_against_verbs("ler", {"^", "^re", "tres"}), forms = { pres_2s = "lês", pres_3s = "lê", pres_2p = "ledes", pres_3p = {"leem", VAR_SUPERSEDED .. "lêem"}, pres1_and_sub = "lei", irreg = true, }	},	{		-- morrer, desmorrer match = "morrer", forms = {short_pp = "morto"} },	{		-- doer, moer/remoer, roer/corroer, soer match = "oer", forms = { pres_1s = function(base, prefix) return prefix ~= "s" and {"oo", VAR_SUPERSEDED .. "ôo"} or nil end, pres_2s = "óis", pres_3s = "ói", -- impf -ía etc., pret_1s -oí and pp -oído handled automatically in combine_stem_ending only3sp = function(base, prefix) return prefix == "d" end, no_pres1_and_sub = function(base, prefix) return prefix == "s" end, irreg = true, }	},	{		-- perder match = "perder", forms = { -- use 'perqu' because we're in a front environment; if we use 'perc', we'll get '#perço' pres1_and_sub = "perqu", irreg = true, }	},	{		-- poder match = "poder", forms = { pres1_and_sub = "poss", pret = "pudé", pret_1s = "pude", pret_3s = "pôde", pret_conj = "irreg", irreg = true, }	},	{		-- prazer, aprazer, comprazer, desprazer match = "prazer", forms = { pres_3s = "praz", pret = "prouvé", pret_1s = "prouve", pret_3s = "prouve", pret_conj = "irreg", only3sp = function(base, prefix) return not prefix:find("com$") end, irreg = true, }	},	-- prover below, just below ver {		-- requerer; must precede querer match = "requerer", forms = { -- old module claims alt pres_3s 'requere'; not in Priberam, Infopédia or conjugacao.com.br			pres_3s = "requer", pres1_and_sub = "requeir", imp_2s = {{form = "requere", footnotes = {"[Brazil only]"}}, "requer"}, -- per Priberam -- regular preterite, unlike querer irreg = true, }	},	{		-- querer, desquerer, malquerer match = "querer", forms = { -- old module claims alt pres_3s 'quere'; not in Priberam, Infopédia or conjugacao.com.br			pres_1s = "quero", pres_3s = "quer", pres1_and_sub = "queir", -- only for subjunctive as we override pres_1s pret = "quisé", pret_1s = "quis", pret_3s = "quis", pret_conj = "irreg", imp_2s = {{form = "quere", footnotes = {"[Brazil only]"}}, {form = "quer", footnotes = {"[Brazil only]"}}}, -- per Priberam irreg = true, }	},	{		match = "reaver", forms = { no_pres_stressed = true, pret = "reouvé", pret_conj = "irreg", pret_1s = "reouve", pret_3s = "reouve", irreg = true, }	},	{		-- saber, resaber match = "saber", forms = { pres_1s = "sei", pres1_and_sub = "saib", -- only for subjunctive as we override pres_1s pret = "soubé", pret_1s = "soube", pret_3s = "soube", pret_conj = "irreg", irreg = true, }	},	{		-- escrever/reescrever, circunscrever, descrever/redescrever, inscrever, prescrever, proscrever, subscrever, -- transcrever, others? match = "screver", forms = { pp = "scrito", irreg = true, }	},	{		-- suspender match = "suspender", forms = {short_pp = "suspenso"} },	{		match = "^ser", forms = { pres_1s = "sou", pres_2s = "és", pres_3s = "é", pres_1p = "somos", pres_2p = "sois", pres_3p = "são", pres1_and_sub = "sej", -- only for subjunctive as we override pres_1s full_impf = "er", impf_1p = "éramos", impf_2p = "éreis", pret = "fô", pret_1s = "fui", pret_3s = "foi", pret_conj = "irreg", imp_2s = "sê", imp_2p = "sede", pp_inv = true, irreg = true, }	},	{		-- We want to match abster, conter, deter, etc. but not abater, cometer, etc. No way to avoid listing each verb. match = match_against_verbs("ter", {"abs", "^a", "con", "de", "entre", "man", "ob", "^re", "sus", "^"}), forms = { pres_2s = function(base, prefix) return prefix == "" and "tens" or "téns" end, pres_3s = function(base, prefix) return prefix == "" and "tem" or "tém" end, pres_2p = "tendes", pres_3p = "têm", pres1_and_sub = "tenh", full_impf = "tinh", impf_1p = "tínhamos", impf_2p = "tínheis", pret = "tivé", pret_1s = "tive", pret_3s = "teve", pret_conj = "irreg", irreg = true, }	},	{		match = "trazer", forms = { -- use 'tragu' because we're in a front environment; if we use 'trag', we'll get '#trajo' pres1_and_sub = "tragu", pres_3s = "traz", pret = "trouxé", pret_1s = "trouxe", pret_3s = "trouxe", pret_conj = "irreg", fut = "trar", irreg = true, }	},	{		-- valer, desvaler, equivaler match = "valer", forms = {pres1_and_sub = "valh"} },	{		-- We want to match antever etc. but not absolver, atrever etc. No way to avoid listing each verb. match = match_against_verbs("ver", {"ante", "entre", "pre", "^re", "^"}), forms = { pres_2s = "vês", pres_3s = "vê", pres_2p = "vedes", pres_3p = {"veem", VAR_SUPERSEDED .. "vêem"}, pres1_and_sub = "vej", pret = "ví", pret_1s = "vi", pret_3s = "viu", pret_conj = "irreg", pp = "visto", irreg = true, }	},	{		-- prover and desprover have regular preterite and past participle match = "prover", forms = { pres_2s = "provês", pres_3s = "provê", pres_2p = "provedes", pres_3p = {"proveem", VAR_SUPERSEDED .. "provêem"}, pres1_and_sub = "provej", irreg = true, }	},	{		-- Only envolver, revolver. Not volver, desenvolver, devolver, evolver, etc.		match = match_against_verbs("volver", {"^en", "^re"}), forms = {short_pp = "volto"}, },

--                                            -ir                                        --

-- Verbs not needing entries here: --	-- abolir: per Priberam:  for Brazil, use  for Portugal -- barrir: use -- carpir, colorir, demolir: use  -- descolorir: per Priberam:  for Brazil, use <no_pres1_and_sub> for Portugal -- delir, espavorir, falir, florir, remir, renhir: use <no_pres_stressed> -- empedernir: per Priberam: <no_pres_stressed> for Brazil, use <i-e> for Portugal -- transir: per Priberam: <no_pres_stressed> for Brazil, regular for Portugal -- aspergir, despir, flectir/deflectir/genuflectir/genufletir/reflectir/refletir, mentir/desmentir, --  sentir/assentir/consentir/dissentir/pressentir/ressentir, convergir/divergir, aderir/adherir, --  ferir/auferir/conferir/deferir/desferir/diferir/differir/inferir/interferir/preferir/proferir/referir/transferir, --  gerir/digerir/ingerir/sugerir, preterir, competir/repetir, servir, advertir/animadvertir/divertir, --  vestir/investir/revestir/travestir, seguir/conseguir/desconseguir/desseguir/perseguir/prosseguir: use <i-e> -- inerir: use <i-e> (per Infopédia, and per Priberam for Brazil), use <i-e.only3sp> (per Priberam for Portugal) -- compelir/expelir/impelir/repelir: per Priberam: use <i-e> for Brazil, <no_pres1_and_sub> for Portugal (Infopédia	--  says <i-e>); NOTE: old module claims short_pp 'repulso' but none of Priberam, Infopédia and conjugacao.com.br agree -- dormir, engolir, tossir, subir, acudir/sacudir, fugir, sumir/consumir (NOT assumir/presumir/resumir): use <u-o> -- polir/repolir (claimed in old module to have no pres stressed, but Priberam disagrees for both Brazil and	--  Portugal; Infopédia lists repolir as completely regular and not like polir, but I think that's an error): use --  	-- premir: per Priberam: use <no_pres1_and_sub> for Brazil, <i> for Portugal (for Portugal, Priberam says	--   primo/primes/prime, while Infopédia says primo/premes/preme; Priberam is probably more reliable) -- extorquir/retorquir use <no_pres1_and_sub> for Brazil, <u-o,u> for Portugal -- agredir/progredir/regredir/transgredir: use <i> -- denegrir, prevenir: use <i> -- eclodir: per Priberam: regular in Brazil, <u-o.only3sp> in Portugal (Infopédia says regular) -- cerzir: per Priberam: use <i> for Brazil, use <i-e> for Portugal (Infopédia says <i-e,i>) -- cergir: per Priberam: use <i-e> for Brazil, no conjugation given for Portugal (Infopédia says <i-e>) -- proibir/coibir: use <í> -- reunir: use <ú> -- parir/malparir: use <no_pres_stressed> (old module had pres_1s = {paro (1_defective), pairo (1_obsolete_alt)},	--  pres_2s = pares, pres_3s = pare, and subjunctive stem par- or pair-, but both Priberam and Infopédia agree	--   in these verbs being no_pres_stressed) -- explodir/implodir: use <u-o> (claimed in old module to be <+,u-o> but neither Priberam nor Infopédia agree) --	-- -cir alternations (aducir, ressarcir): automatically handled in combine_stem_ending -- -gir alternations (agir, dirigir, exigir): automatically handled in combine_stem_ending -- -guir alternations (e.g. conseguir): automatically handled in combine_stem_ending -- -quir alternations (e.g. extorquir): automatically handled in combine_stem_ending

{		-- verbs in -air (cair, sair, trair and derivatives: decair/descair/recair, sobres(s)air,		-- abstrair/atrair/contrair/distrair/extrair/protrair/retrair/subtrair) match = "air", forms = { pres1_and_sub = "ai", pres_2s = "ais", pres_3s = "ai", -- all occurrences of accented í in endings handled in combine_stem_ending irreg = true, }	},	{		-- abrir/desabrir/reabrir match = "abrir", forms = {pp = "aberto"} },	{		-- cobrir/descobrir/encobrir/recobrir/redescobrir match = "cobrir", forms = {vowel_alt = "u-o", pp = "coberto"} },	{		-- conduzir, produzir, reduzir, traduzir, etc.; luzir, reluzir, tremeluzir match = "uzir", forms = { pres_3s = "uz", imp_2s = {"uz", "uze"}, -- per Infopédia irreg = true, }	},	{		-- pedir, desimpedir, despedir, espedir, expedir, impedir -- medir -- comedir (per Priberam, no_pres_stressed in Brazil) match = match_against_verbs("edir", {"m", "p"}), forms = { pres1_and_sub = "eç", irreg = true, }	},	{		-- frigir match = "frigir", forms = {vowel_alt = "i-e", short_pp = "frito"}, },		{		-- inserir match = "inserir", forms = {vowel_alt = "i-e", short_pp = {form = "inserto", footnotes = {"[Portugal only]"}}}, },	{		-- ir		match = "^ir", forms = { pres_1s = "vou", pres_2s = "vais", pres_3s = "vai", pres_1p = "vamos", pres_2p = "ides", pres_3p = "vão", pres_sub_1s = "vá", pres_sub_2s = "vás", pres_sub_3s = "vá", pres_sub_1p = "vamos", pres_sub_2p = "vades", pres_sub_3p = "vão", pret = "fô", pret_1s = "fui", pret_3s = "foi", pret_conj = "irreg", irreg = true, }	},	{		-- emergir, imergir, submergir match = "mergir", forms = {vowel_alt = {"i-e", "+"}, short_pp = "merso"}, },	{		match = "ouvir", forms = { pres1_and_sub = {"ouç", "oiç"}, irreg = true, }	},	{		-- exprimir, imprimir, comprimir (but not descomprimir per Priberam), deprimir, oprimir/opprimir (but not reprimir,		-- suprimir/supprimir per Priberam) match = match_against_verbs("primir", {"^com", "ex", "im", "de", "^o", "op"}), forms = {short_pp = "presso"} },	{		-- rir, sorrir match = match_against_verbs("rir", {"^", "sor"}), forms = { pres_2s = "ris", pres_3s = "ri", pres_2p = "rides", pres_3p = "riem", pres1_and_sub = "ri", irreg = true, }	},	{		-- distinguir, extinguir match = "tinguir", forms = { short_pp = "tinto", -- gu/g alternations handled in combine_stem_ending }	},	{		-- delinquir, arguir/redarguir -- NOTE: The following is based on delinquir, with arguir/redarguir by parallelism. -- In Priberam, delinquir and arguir are exactly parallel, but in Infopédia they aren't; only delinquir has -- alternatives like 'delínques'. I assume this is because forms like 'delínques' are Brazilian and -- Infopédia is from Portugal, so their coverage of Brazilian forms may be inconsistent. match = match_against_verbs("uir", {"delinq", "arg"}), forms = { -- use 'ü' because we're in a front environment; if we use 'u', we'll get '#delinco', '#argo' pres1_and_sub = {{form = AC .. "ü", footnotes = {"[Brazil]"}}, {form = "ü", footnotes = {"[Portugal]"}}}, -- FIXME: verify. This is by partial parallelism with the present subjunctive of verbs in -quar (also a			-- front environment). Infopédia has 'delinquis ou delínques' and Priberam has 'delinqúis'. pres_2s = { {form = AC .. "ues", footnotes = {"[Brazil]"}}, {form = "uis", footnotes = {"[Portugal]"}}, -- This form should occur only with an infinitive 'delinqüir' etc. -- {form = AC .. VAR_SUPERSEDED .. "ües", footnotes = {"[Brazil]"}}, {form = VAR_SUPERSEDED .. "úis", footnotes = {"[Portugal]"}}, },			-- Same as previous. pres_3s = { {form = AC .. "ue", footnotes = {"[Brazil]"}}, {form = "ui", footnotes = {"[Portugal]"}}, -- This form should occur only with an infinitive 'delinqüir' etc. -- {form = AC .. VAR_SUPERSEDED .. "üe", footnotes = {"[Brazil]"}}, {form = VAR_SUPERSEDED .. "úi", footnotes = {"[Portugal]"}}, },			-- Infopédia has 'delinquem ou delínquem' and Priberam has 'delinqúem'. pres_3p = { {form = AC .. "uem", footnotes = {"[Brazil]"}}, {form = "uem", footnotes = {"[Portugal]"}}, -- This form should occur only with an infinitive 'delinqüir' etc. -- {form = AC .. VAR_SUPERSEDED .. "üem", footnotes = {"[Brazil]"}}, {form = VAR_SUPERSEDED .. "úem", footnotes = {"[Portugal]"}}, },			-- FIXME: The old module also had several other alternative forms (given as [123]_alt, not identified as			-- obsolete): -- impf: delinquia/delinquía, delinquias/delinquías, delinquia/delinquía, delinquíamos, delinquíeis, delinquiam/delinquíam -- plup: delinquira/delinquíra, delinquiras/delinquíras, delinquira/delinquíra, delinquíramos, delinquíreis, delinquiram/delinquíram -- pres_1p = delinquimos/delinquímos, pres_2p = delinquis/delinquís -- pret = delinqui/delinquí, delinquiste/delinquíste, delinquiu, delinquimos/delinquímos, delinquistes/delinquístes, delinquiram/delinquíram -- pers_inf = delinquir, delinquires, delinquir, delinquirmos, delinquirdes, delinquirem/delinquírem -- fut_sub = delinquir, delinquires, delinquir, delinquirmos, delinquirdes, delinquirem/delinquírem --			-- None of these alternative forms can be found in the Infopédia, Priberam, Collins or Reverso conjugation -- tables, so their status is unclear, and I have omitted them. }	},	{		-- verbs in -truir (construir, destruir, reconstruir) but not obstruir/desobstruir, instruir, which are handled -- by the default -uir handler below match = match_against_verbs("struir", {"con", "de"}), forms = { pres_2s = {"stróis", "struis"}, pres_3s = {"strói", "strui"}, pres_3p = {"stroem", "struem"}, -- all occurrences of accented í in endings handled in combine_stem_ending irreg = true, }	},	{		-- verbs in -cluir (concluir, excluir, incluir): like -uir but has short_pp concluso etc. in Brazil match = "cluir", forms = { pres_2s = "cluis", pres_3s = "clui", -- all occurrences of accented í in endings handled in combine_stem_ending short_pp = {form = "cluso", footnotes = {"[Brazil only]"}}, irreg = true, }	},	{		-- puir, ruir: like -uir but defective in pres_1s, all pres sub match = match_against_verbs("uir", {"^p", "^r"}), forms = { pres_2s = "uis", pres_3s = "ui", -- all occurrences of accented í in endings handled in combine_stem_ending no_pres1_and_sub = true, irreg = true, }	},	{		-- remaining verbs in -uir (concluir/excluir/incluir/concruir/concruyr, abluir/diluir, afluir/fluir/influir,		-- aluir, anuir, atribuir/contribuir/distribuir/redistribuir/retribuir/substituir, coevoluir/evoluir,		-- constituir/destituir/instituir/reconstituir/restituir, derruir, diminuir, estatuir, fruir/usufruir, imbuir,		-- imiscuir, poluir, possuir, pruir		-- FIXME: old module lists short pp incluso for incluir that can't be verified, ask about this		-- FIXME: handle -uyr verbs?		match = function(verb)			-- Don't match -guir verbs (e.g. seguir, conseguir) or -quir verbs (e.g. extorquir)			if verb:find("guir") or verb:find("quir") then				return nil			else				return match_against_verbs("uir", {""})(verb)			end		end,		forms = {			pres_2s = "uis", pres_3s = "ui",			-- all occurrences of accented í in endings handled in combine_stem_ending			irreg = true,		}	},	{ -- We want to match advir, convir, devir, etc. but not ouvir, servir, etc. No way to avoid listing each verb. match = match_against_verbs("vir", {"ad", "^a", "con", "contra", "de", "^desa", "inter", "pro", "^re", "sobre", "^"}), forms = { pres_2s = function(base, prefix) return prefix == "" and "vens" or "véns" end, pres_3s = function(base, prefix) return prefix == "" and "vem" or "vém" end, pres_2p = "vindes", pres_3p = "vêm", pres1_and_sub = "venh", full_impf = "vinh", impf_1p = "vínhamos", impf_2p = "vínheis", pret = "vié", pret_1s = "vim", pret_3s = "veio", pret_conj = "irreg", pp = "vindo", irreg = true, }	},

--                                           misc                                        --

{		-- pôr, antepor, apor, compor/decompor/descompor, contrapor, depor, dispor, expor, impor, interpor, justapor, -- opor, pospor, propor, repor, sobrepor, supor/pressupor, transpor, superseded forms like decompôr, others? match = "p[oô]r", forms = { pres1_and_sub = "ponh", pres_2s = "pões", pres_3s = "põe", pres_1p = "pomos", pres_2p = "pondes", pres_3p = "põem", full_impf = "punh", impf_1p = "púnhamos", impf_2p = "púnheis", pret = "pusé", pret_1s = "pus", pret_3s = "pôs", pret_conj = "irreg", pers_inf = "po", gerund = "pondo", pp = "posto", irreg = true, }	}, }

local function skip_slot(base, slot, allow_overrides) if not allow_overrides and (base.basic_overrides[slot] or		base.refl and base.basic_reflexive_only_overrides[slot]) then -- Skip any slots for which there are overrides. return true end

if base.only3s and (slot:find("^pp_f") or slot:find("^pp_mp")) then -- diluviar, atardecer, neviscar; impersonal verbs have only masc sing pp		return true end

if not slot:find("[123]") then -- Don't skip non-personal slots. return false end

if base.nofinite then return true end

if (base.only3s or base.only3sp or base.only3p) and (slot:find("^imp_") or slot:find("^neg_imp_")) then return true end

if base.only3s and not slot:find("3s") then -- diluviar, atardecer, neviscar return true end

if base.only3sp and not slot:find("3[sp]") then -- atañer, concernir return true end

if base.only3p and not slot:find("3p") then -- caer cuatro gotas, caer chuzos de punta, entrarle los siete males return true end

return false end

-- Apply vowel alternations to stem. local function apply_vowel_alternations(stem, alternations) local alternation_stems = {} local saw_pres1_and_sub = false local saw_pres_stressed = false

-- Process alternations other than +. for _, altobj in ipairs(alternations) do		local alt = altobj.form local pres1_and_sub, pres_stressed, err -- Treat final -gu, -qu as a consonant, so the previous vowel can alternate (e.g. conseguir -> consigo). -- This means a verb in -guar can't have a u-ú alternation but I don't think there are any verbs like that. stem = rsub(stem, "([gq])u$", "%1" .. TEMPC1) if alt == "+" then -- do nothing yet elseif alt == "ei" then local before_last_vowel = rmatch(stem, "^(.*)i$") if not before_last_vowel then err = "stem should end in -i" else pres1_and_sub = nil pres_stressed = before_last_vowel .. "ei" end else local before_last_vowel, last_vowel, after_last_vowel = rmatch(stem, "^(.*)(" .. V .. ")(.-[ui])$") if not before_last_vowel then before_last_vowel, last_vowel, after_last_vowel = rmatch(stem, "^(.*)(" .. V .. ")(.-)$")			end if alt == "i-e" then if last_vowel == "e" or last_vowel == "i" then pres1_and_sub = before_last_vowel .. "i" .. after_last_vowel if last_vowel == "i" then pres_stressed = before_last_vowel .. "e" .. after_last_vowel end else err = "should have -e- or -i- as the last vowel" end elseif alt == "i" then if last_vowel == "e" then pres1_and_sub = before_last_vowel .. "i" .. after_last_vowel pres_stressed = pres1_and_sub else err = "should have -e- as the last vowel" end elseif alt == "u-o" then if last_vowel == "o" or last_vowel == "u" then pres1_and_sub = before_last_vowel .. "u" .. after_last_vowel if last_vowel == "u" then pres_stressed = before_last_vowel .. "o" .. after_last_vowel end else err = "should have -o- or -u- as the last vowel" end elseif alt == "u" then if last_vowel == "o" then pres1_and_sub = before_last_vowel .. "u" .. after_last_vowel pres_stressed = pres1_and_sub else err = "should have -o- as the last vowel" end elseif alt == "í" then if last_vowel == "i" then pres_stressed = before_last_vowel .. "í" .. after_last_vowel else err = "should have -i- as the last vowel" end elseif alt == "ú" then if last_vowel == "u" then pres_stressed = before_last_vowel .. "ú" .. after_last_vowel else err = "should have -u- as the last vowel" end else error("Internal error: Unrecognized vowel alternation '" .. alt .. "'") end end if pres1_and_sub then pres1_and_sub = {form = pres1_and_sub:gsub(TEMPC1, "u"), footnotes = altobj.footnotes} saw_pres1_and_sub = true end if pres_stressed then pres_stressed = {form = pres_stressed:gsub(TEMPC1, "u"), footnotes = altobj.footnotes} saw_pres_stressed = true end table.insert(alternation_stems, {			altobj = altobj,			pres1_and_sub = pres1_and_sub,			pres_stressed = pres_stressed,			err = err		}) end

-- Now do +. We check to see which stems are used by other alternations and specify those so any footnotes are -- properly attached. for _, alternation_stem in ipairs(alternation_stems) do		if alternation_stem.altobj.form == "+" then local stemobj = {form = stem, footnotes = alternation_stem.altobj.footnotes} alternation_stem.pres1_and_sub = saw_pres1_and_sub and stemobj or nil alternation_stem.pres_stressed = saw_pres_stressed and stemobj or nil end end

return alternation_stems end

-- Add the `stem` to the `ending` for the given `slot` and apply any phonetic modifications. -- WARNING: This function is written very carefully; changes to it can easily have unintended consequences. local function combine_stem_ending(base, slot, prefix, stem, ending, dont_include_prefix) -- If the stem begins with an acute accent, this is a signal to move the accent onto the last vowel of the prefix. -- Cf. míngua of minguar. if stem:find("^" .. AC) then stem = rsub(stem, "^" .. AC, "") if dont_include_prefix then error("Internal error: Can't handle acute accent at beginning of stem if dont_include_prefix is given") end prefix = rsub(prefix, "([aeiouyAEIOUY])([^aeiouyAEIOUY]*)$", "%1" .. AC .. "%2") end

-- Use the full stem for checking for -gui ending and such, because 'stem' is just 'u' for arguir, -- delinquir. local full_stem = prefix .. stem -- Include the prefix in the stem unless dont_include_prefix is given (used for the past participle stem). if not dont_include_prefix then stem = prefix .. stem end

-- If the ending begins with a double asterisk, this is a signal to conditionally delete the accent on the last letter -- of the stem. "Conditionally" means we don't do it if the last two letters would form a diphthong without the accent -- on the second one (e.g. in sair, with stem 'saí'); but as an exception, we do delete the accent in stems -- ending in -guí, -quí (e.g. in conseguir) because in this case the ui isn't a diphthong. if ending:find("^%*%*") then ending = rsub(ending, "^%*%*", "") if rfind(full_stem, "[gq]uí$") or not rfind(full_stem, V .. "[íú]$") then stem = com.remove_final_accent(stem) end end

-- If the ending begins with an asterisk, this is a signal to delete the accent on the last letter of the stem. -- E.g. fizé -> fizermos. Unlike for **, this removal is unconditional, so we get e.g. 'sairmos' not #'saírmos'. if ending:find("^%*") then ending = rsub(ending, "^%*", "") stem = com.remove_final_accent(stem) end

-- If ending begins with i, it must get an accent after an unstressed vowel (in some but not all cases) to prevent -- the two merging into a diphthong: -- * cair -> -- *  pres: caímos, caís; -- *  impf: all forms (caí-); -- *  pret: caí, caíste (but not caiu), caímos, caístes, caíram; -- *  plup: all forms (caír-); -- *  impf_sub: all forms (caíss-); -- *  fut_sub: caíres, caírem (but not cair, cairmos, cairdes) -- *  pp: caído (but not gerund caindo) -- * atribuir, other verbs in -uir -> same pattern as for cair etc.	-- * roer -> -- *  pret: roí -- *  impf: all forms (roí-) -- *  pp: roído if ending:find("^i") and full_stem:find("[aeiou]$") and not full_stem:find("[gq]u$") and ending ~= "ir" and ending ~= "iu" and ending ~= "indo" and not ending:find("^ir[md]") then ending = ending:gsub("^i", "í") end

-- Spelling changes in the stem; it depends on whether the stem given is the pre-front-vowel or -- pre-back-vowel variant, as indicated by `frontback`. We want these front-back spelling changes to happen -- between stem and ending, not between prefix and stem; the prefix may not have the same "front/backness" -- as the stem. local is_front = rfind(ending, "^[eiéí]") if base.frontback == "front" and not is_front then stem = stem:gsub("c$", "ç") -- conhecer -> conheço, vencer -> venço, descer -> desço stem = stem:gsub("g$", "j") -- proteger -> protejo, fugir -> fujo stem = stem:gsub("gu$", "g") -- distinguir -> distingo, conseguir -> consigo stem = stem:gsub("qu$", "c") -- extorquir -> exturco stem = stem:gsub("([gq])ü$", "%1u") -- argüir (superseded) -> arguo, delinqüir (superseded) -> delinquo elseif base.frontback == "back" and is_front then -- The following changes are all superseded so we don't do them: -- averiguar -> averigüei, minguar -> mingüei; antiquar -> antiqüei, apropinquar -> apropinqüei -- stem = stem:gsub("([gq])u$", "%1ü") stem = stem:gsub("g$", "gu") -- cargar -> carguei, apagar -> apaguei stem = stem:gsub("c$", "qu") -- marcar -> marquei stem = stem:gsub("ç$", "c") -- começar -> comecei -- j does not go to g here; desejar -> deseje not #desege end

return stem .. ending end

local function add3(base, slot, stems, endings, footnotes, allow_overrides) if skip_slot(base, slot, allow_overrides) then return end

local function do_combine_stem_ending(stem, ending) return combine_stem_ending(base, slot, base.prefix, stem, ending) end

iut.add_forms(base.forms, slot, stems, endings, do_combine_stem_ending, nil, nil, footnotes) end

local function insert_form(base, slot, form) if not skip_slot(base, slot) then iut.insert_form(base.forms, slot, form) end end

local function insert_forms(base, slot, forms) if not skip_slot(base, slot) then iut.insert_forms(base.forms, slot, forms) end end

local function add_single_stem_tense(base, slot_pref, stems, s1, s2, s3, p1, p2, p3) local function addit(slot, ending) add3(base, slot_pref .. "_" .. slot, stems, ending) end addit("1s", s1) addit("2s", s2) addit("3s", s3) addit("1p", p1) addit("2p", p2) addit("3p", p3) end

local function construct_stems(base, vowel_alt) local stems = {} stems.pres_unstressed = base.stems.pres_unstressed or base.inf_stem stems.pres_stressed = -- If no_pres_stressed given, pres_stressed stem should be empty so no forms are generated. base.no_pres_stressed and {} or		base.stems.pres_stressed or		vowel_alt.pres_stressed or		base.inf_stem stems.pres1_and_sub = -- If no_pres_stressed given, the entire subjunctive is missing. base.no_pres_stressed and {} or -- If no_pres1_and_sub given, pres1 and entire subjunctive are missing. base.no_pres1_and_sub and {} or		base.stems.pres1_and_sub or		vowel_alt.pres1_and_sub or		nil stems.pres1 = base.stems.pres1 or stems.pres1_and_sub or stems.pres_stressed stems.impf = base.stems.impf or base.inf_stem stems.full_impf = base.stems.full_impf stems.pret_base = base.stems.pret_base or base.inf_stem stems.pret = base.stems.pret or iut.map_forms(iut.convert_to_general_list_form(stems.pret_base), function(form)		return form .. base.conj_vowel end) stems.pret_conj = base.stems.pret_conj or base.conj stems.fut = base.stems.fut or base.inf_stem .. base.conj stems.cond = base.stems.cond or stems.fut stems.pres_sub_stressed = base.stems.pres_sub_stressed or stems.pres1 stems.pres_sub_unstressed = base.stems.pres_sub_unstressed or stems.pres1_and_sub or stems.pres_unstressed stems.sub_conj = base.stems.sub_conj or base.conj stems.plup = base.stems.plup or stems.pret stems.impf_sub = base.stems.impf_sub or stems.pret stems.fut_sub = base.stems.fut_sub or stems.pret stems.pers_inf = base.stems.pers_inf or base.inf_stem .. base.conj_vowel stems.pp = base.stems.pp or base.conj == "ar" and combine_stem_ending(base, "pp_ms", base.prefix, base.inf_stem, "ado", "dont include prefix") or -- use combine_stem_ending esp. so we get roído, caído, etc.		combine_stem_ending(base, "pp_ms", base.prefix, base.inf_stem, "ido", "dont include prefix") stems.pp_ms = stems.pp	local function masc_to_fem(form) if rfind(form, "o$") then return rsub(form, "o$", "a") else return form end end stems.pp_fs = iut.map_forms(iut.convert_to_general_list_form(stems.pp_ms), masc_to_fem) if base.stems.short_pp then stems.short_pp_ms = base.stems.short_pp stems.short_pp_fs = iut.map_forms(iut.convert_to_general_list_form(stems.short_pp_ms), masc_to_fem) end base.this_stems = stems end

local function add_present_indic(base) local stems = base.this_stems local function addit(slot, stems, ending) add3(base, "pres_" .. slot, stems, ending) end local s2, s3, p1, p2, p3	if base.conj == "ar" then s2, s3, p1, p2, p3 = "as", "a", "amos", "ais", "am" elseif base.conj == "er" or base.conj == "or" then -- verbs in -por have the present overridden s2, s3, p1, p2, p3 = "es", "e", "emos", "eis", "em" elseif base.conj == "ir" then s2, s3, p1, p2, p3 = "es", "e", "imos", "is", "em" else error("Internal error: Unrecognized conjugation " .. base.conj) end

addit("1s", stems.pres1, "o") addit("2s", stems.pres_stressed, s2) addit("3s", stems.pres_stressed, s3) addit("1p", stems.pres_unstressed, p1) addit("2p", stems.pres_unstressed, p2) addit("3p", stems.pres_stressed, p3) end

local function add_present_subj(base) local stems = base.this_stems local function addit(slot, stems, ending) add3(base, "pres_sub_" .. slot, stems, ending) end

local s1, s2, s3, p1, p2, p3	if stems.sub_conj == "ar" then s1, s2, s3, p1, p2, p3 = "e", "es", "e", "emos", "eis", "em" else s1, s2, s3, p1, p2, p3 = "a", "as", "a", "amos", "ais", "am" end

addit("1s", stems.pres_sub_stressed, s1) addit("2s", stems.pres_sub_stressed, s2) addit("3s", stems.pres_sub_stressed, s3) addit("1p", stems.pres_sub_unstressed, p1) addit("2p", stems.pres_sub_unstressed, p2) addit("3p", stems.pres_sub_stressed, p3) end

local function add_finite_non_present(base) local stems = base.this_stems local function add_tense(slot, stem, s1, s2, s3, p1, p2, p3) add_single_stem_tense(base, slot, stem, s1, s2, s3, p1, p2, p3) end

if stems.full_impf then -- An override needs to be supplied for the impf_1p and impf_2p due to the written accent on the stem. add_tense("impf", stems.full_impf, "a", "as", "a", {}, {}, "am") elseif base.conj == "ar" then add_tense("impf", stems.impf, "ava", "avas", "ava", "ávamos", "áveis", "avam") else add_tense("impf", stems.impf, "ia", "ias", "ia", "íamos", "íeis", "iam") end

-- * at the beginning of the ending means to remove a final accent from the preterite stem. if stems.pret_conj == "irreg" then add_tense("pret", stems.pret, {}, "*ste", {}, "*mos", "*stes", "*ram") elseif stems.pret_conj == "ar" then add_tense("pret", stems.pret_base, "ei", "aste", "ou",			{{form = VAR_BR .. "amos", footnotes = {"[Brazil]"}}, {form = VAR_PT .. "ámos", footnotes = {"[Portugal]"}}}, "astes", "aram") elseif stems.pret_conj == "er" then add_tense("pret", stems.pret_base, "i", "este", "eu", "emos", "estes", "eram") else add_tense("pret", stems.pret_base, "i", "iste", "iu", "imos", "istes", "iram") end

-- * at the beginning of the ending means to remove a final accent from the stem. -- ** is similar but is "conditional" on a consonant preceding the final vowel. add_tense("plup", stems.plup, "**ra", "**ras", "**ra", "ramos", "reis", "**ram") add_tense("impf_sub", stems.impf_sub, "**sse", "**sses", "**sse", "ssemos", "sseis", "**ssem") add_tense("fut_sub", stems.fut_sub, "*r", "**res", "*r", "*rmos", "*rdes", "**rem") local mark = TEMP_MESOCLITIC_INSERTION_POINT add_tense("fut", stems.fut, mark .. "ei", mark .. "ás", mark .. "á", mark .. "emos", mark .. "eis", mark .. "ão") add_tense("cond", stems.cond, mark .. "ia", mark .. "ias", mark .. "ia", mark .. "íamos", mark .. "íeis", mark .. "iam") -- Different stems for different parts of the personal infinitive to correctly handle forms of sair and pôr. add_tense("pers_inf", base.non_prefixed_verb, "", {}, "", {}, {}, {}) add_tense("pers_inf", stems.pers_inf, {}, "**res", {}, "*rmos", "*rdes", "**rem") end

local function add_non_finite_forms(base) local stems = base.this_stems local function addit(slot, stems, ending, footnotes) add3(base, slot, stems, ending, footnotes) end

insert_form(base, "infinitive", {form = base.verb}) -- Also insert "infinitive + reflexive pronoun" combinations if we're handling a reflexive verb. See comment below for -- "gerund + reflexive pronoun" combinations. if base.refl then for _, persnum in ipairs(person_number_list) do insert_form(base, "infinitive_" .. persnum, {form = base.verb}) end end -- verbs in -por have the gerund overridden local ger_ending = base.conj == "ar" and "ando" or base.conj == "er" and "endo" or "indo" addit("gerund", stems.pres_unstressed, ger_ending) -- Also insert "gerund + reflexive pronoun" combinations if we're handling a reflexive verb. We insert exactly the same -- form as for the bare gerund; later on in add_reflexive_or_fixed_clitic_to_forms, we add the appropriate clitic -- pronouns. It's important not to do this for non-reflexive verbs, because in that case, the clitic pronouns won't be	-- added, and will wrongly consider all these combinations as possible inflections of the bare -- gerund. Thanks to User:JeffDoozan for this bug fix. if base.refl then for _, persnum in ipairs(person_number_list) do addit("gerund_" .. persnum, stems.pres_unstressed, ger_ending) end end -- Skip the long/short past participle footnotes if called from so they don't show in the headword. local long_pp_footnotes = stems.short_pp_ms and not base.alternant_multiword_spec.from_headword and {long_pp_footnote} or nil addit("pp_ms", stems.pp_ms, "", long_pp_footnotes) if not base.pp_inv then addit("pp_fs", stems.pp_fs, "", long_pp_footnotes) addit("pp_mp", stems.pp_ms, "s", long_pp_footnotes) addit("pp_fp", stems.pp_fs, "s", long_pp_footnotes) end if stems.short_pp_ms then local short_pp_footnotes = stems.short_pp_ms and not base.alternant_multiword_spec.from_headword and {short_pp_footnote} or nil addit("short_pp_ms", stems.short_pp_ms, "", short_pp_footnotes) if not base.pp_inv then addit("short_pp_fs", stems.short_pp_fs, "", short_pp_footnotes) addit("short_pp_mp", stems.short_pp_ms, "s", short_pp_footnotes) addit("short_pp_fp", stems.short_pp_fs, "s", short_pp_footnotes) end end end

local function copy_forms_to_imperatives(base) -- Copy pres3s to imperative since they are almost always the same. insert_forms(base, "imp_2s", iut.map_forms(base.forms.pres_3s, function(form) return form end)) if not skip_slot(base, "imp_2p") then -- Copy pres2p to imperative 2p minus -s since they are almost always the same. -- But not if there's an override, to avoid possibly throwing an error. insert_forms(base, "imp_2p", iut.map_forms(base.forms.pres_2p, function(form) if not form:find("s$") then error("Can't derive second-person plural imperative from second-person plural present indicative " ..					"because form '" .. form .. "' doesn't end in -s") end return rsub(form, "s$", "") end)) end -- Copy subjunctives to imperatives, unless there's an override for the given slot (as with the imp_1p of ir). for _, persnum in ipairs({"3s", "1p", "3p"}) do local from = "pres_sub_" .. persnum local to = "imp_" .. persnum insert_forms(base, to, iut.map_forms(base.forms[from], function(form) return form end)) end end

local function process_slot_overrides(base, filter_slot, reflexive_only) local overrides = reflexive_only and base.basic_reflexive_only_overrides or base.basic_overrides for slot, forms in pairs(overrides) do		if not filter_slot or filter_slot(slot) then add3(base, slot, forms, "", nil, "allow overrides") end end end

-- Prefix `form` with `clitic`, adding fixed text `between` between them. Add links as appropriate unless the user -- requested no links. Check whether form already has brackets (as will be the case if the form has a fixed clitic). local function prefix_clitic_to_form(base, clitic, between, form) if base.alternant_multiword_spec.args.noautolinkverb then return clitic .. between .. form else local clitic_pref = "" .. clitic .. "" .. between if form:find("%[%[") then return clitic_pref .. form else return clitic_pref .. "" .. form .. "" end end end

-- Add the appropriate clitic pronouns in `clitics` to the forms in `base_slot`. `store_cliticized_form` is a function -- of three arguments (clitic, formobj, cliticized_form) and should store the cliticized form for the specified clitic -- and form object. local function suffix_clitic_to_forms(base, base_slot, clitics, store_cliticized_form) if not base.forms[base_slot] then -- This can happen, e.g. in only3s/only3sp/only3p verbs. return end local autolink = not base.alternant_multiword_spec.args.noautolinkverb for _, formobj in ipairs(base.forms[base_slot]) do		for _, clitic in ipairs(clitics) do			local cliticized_form if formobj.form:find(TEMP_MESOCLITIC_INSERTION_POINT) then -- mesoclisis in future and conditional local infinitive, suffix = rmatch(formobj.form, "^(.*)" .. TEMP_MESOCLITIC_INSERTION_POINT .. "(.*)$") if not infinitive then error("Internal error: Can't find mesoclitic insertion point in slot '" .. base_slot .. "', form '" ..						formobj.form .. "'") end local full_form = infinitive .. suffix if autolink and not infinitive:find("%[%[") then infinitive = "" .. infinitive .. "" end cliticized_form = autolink and infinitive .. "-" .. clitic .. "-" .. suffix .. "" or infinitive .. "-" .. clitic .. "-" .. suffix else local clitic_suffix = autolink and "-" .. clitic .. "" or "-" .. clitic local form_needs_link = autolink and not formobj.form:find("%[%[") if base_slot:find("1p$") then -- Final -s disappears: esbaldávamos + nos -> esbaldávamo-nos, etc.					cliticized_form = formobj.form:gsub("s$", "") if form_needs_link then cliticized_form = "" .. cliticized_form .. "" end else cliticized_form = formobj.form if form_needs_link then cliticized_form = "" .. cliticized_form .. "" end end cliticized_form = cliticized_form .. clitic_suffix end store_cliticized_form(clitic, formobj, cliticized_form) end end end

-- Add a reflexive pronoun or fixed clitic (FIXME: not working), as appropriate to the base forms that were generated. -- `do_joined` means to do only the forms where the pronoun is joined to the end of the form; otherwise, do only the -- forms where it is not joined and precedes the form. local function add_reflexive_or_fixed_clitic_to_forms(base, do_reflexive, do_joined) for _, slotaccel in ipairs(base.alternant_multiword_spec.verb_slots_basic) do		local slot, accel = unpack(slotaccel) local clitic if not do_reflexive then clitic = base.clitic elseif slot:find("[123]") then local persnum = slot:match("^.*_(.-)$") clitic = person_number_to_reflexive_pronoun[persnum] else clitic = "se" end if base.forms[slot] then if do_reflexive and slot:find("^pp_") or slot == "infinitive_linked" then -- do nothing with reflexive past participles or with infinitive linked (handled at the end) elseif slot:find("^neg_imp_") then error("Internal error: Should not have forms set for negative imperative at this stage") else local slot_has_suffixed_clitic = not slot:find("_sub") -- Maybe generate non-reflexive parts and separated syntactic variants for use in. -- See comment in add_slots above `need_special_verb_form_of_slots`. Check for do_joined so we only -- run this code once. if do_reflexive and do_joined and base.alternant_multiword_spec.from_verb_form_of and -- Skip personal variants of infinitives and gerunds so we don't think esbaldando is a -- non-reflexive equivalent of esbaldando-me. not slot:find("infinitive_") and not slot:find("gerund_") then -- Clone the forms because we will be destructively modifying them just below, adding the reflexive -- pronoun. insert_forms(base, slot .. "_non_reflexive", mw.clone(base.forms[slot])) if slot_has_suffixed_clitic then insert_forms(base, slot .. "_variant", iut.map_forms(base.forms[slot], function(form) return prefix_clitic_to_form(base, clitic, " ... ", form) end)) end end if slot_has_suffixed_clitic then if do_joined then suffix_clitic_to_forms(base, slot, {clitic},							function(clitic, formobj, cliticized_form)								formobj.form = cliticized_form							end						) end elseif not do_joined then -- Add clitic as separate word before all other forms. for _, form in ipairs(base.forms[slot]) do						form.form = prefix_clitic_to_form(base, clitic, " ", form.form) end end end end end end

local function handle_infinitive_linked(base) -- Compute linked versions of potential lemma slots, for use in. -- We substitute the original lemma (before removing links) for forms that -- are the same as the lemma, if the original lemma has links. for _, slot in ipairs({"infinitive"}) do insert_forms(base, slot .. "_linked", iut.map_forms(base.forms[slot], function(form) if form == base.lemma and rfind(base.linked_lemma, "%[%[") then return base.linked_lemma else return form end end)) end end

local function generate_negative_imperatives(base) -- Copy subjunctives to negative imperatives, preceded by "não". for _, persnum in ipairs(neg_imp_person_number_list) do local from = "pres_sub_" .. persnum local to = "neg_imp_" .. persnum insert_forms(base, to, iut.map_forms(base.forms[from], function(form) if base.alternant_multiword_spec.args.noautolinkverb then return "não " .. form elseif form:find("%[%[") then -- already linked, e.g. when reflexive return "não " .. form else return "não " .. form .. "" end end)) end end

local function add_missing_links_to_forms(base) -- Any forms without links should get them now. Redundant ones will be stripped later. for slot, forms in pairs(base.forms) do		for _, form in ipairs(forms) do			if not form.form:find("%[%[") then form.form = "" .. form.form .. "" end end end end

-- Remove special characters added to future and conditional forms to indicate mesoclitic insertion points. local function remove_mesoclitic_insertion_points(base) for slot, forms in pairs(base.forms) do		if slot:find("^fut_") or slot:find("^cond_") then for _, form in ipairs(forms) do				form.form = form.form:gsub(TEMP_MESOCLITIC_INSERTION_POINT, "") end end end end

-- If called from, remove superseded forms; otherwise add a footnote indicating they are superseded. local function process_superseded_forms(base) if base.alternant_multiword_spec.from_headword then for slot, forms in pairs(base.forms) do -- As an optimization, check if there are any superseded forms and don't do anything if not. local saw_superseded = false for _, form in ipairs(forms) do				if form.form:find(VAR_SUPERSEDED) then saw_superseded = true break end end if saw_superseded then base.forms[slot] = iut.flatmap_forms(base.forms[slot], function(form)					if form:find(VAR_SUPERSEDED) then						return {}					else						return {form}					end				end) end end else for slot, forms in pairs(base.forms) do			for _, form in ipairs(forms) do				if form.form:find(VAR_SUPERSEDED) then form.footnotes = iut.combine_footnotes(form.footnotes, {"[superseded]"}) end end end end end

local function conjugate_verb(base) for _, vowel_alt in ipairs(base.vowel_alt_stems) do		construct_stems(base, vowel_alt) add_present_indic(base) add_present_subj(base) end add_finite_non_present(base) add_non_finite_forms(base) -- do non-reflexive non-imperative slot overrides process_slot_overrides(base, function(slot)		return not slot:find("^imp_") and not slot:find("^neg_imp_")	end) -- This should happen after process_slot_overrides in case a derived slot is based on an override -- (as with the imp_3s of dar, estar). copy_forms_to_imperatives(base) -- do non-reflexive positive imperative slot overrides process_slot_overrides(base, function(slot)		return slot:find("^imp_")	end) -- We need to add joined reflexives, then joined and non-joined clitics, then non-joined reflexives, so we get -- esbalda-te but não te esbalde. if base.refl then -- This should happen after remove_monosyllabic_accents so the * marking the preservation of monosyllabic -- accents doesn't end up in the middle of a word. add_reflexive_or_fixed_clitic_to_forms(base, "do reflexive", "do joined") process_slot_overrides(base, nil, "do reflexive") -- do reflexive-only slot overrides add_reflexive_or_fixed_clitic_to_forms(base, "do reflexive", false) end -- This should happen after add_reflexive_or_fixed_clitic_to_forms so negative imperatives get the reflexive pronoun -- and clitic in them. generate_negative_imperatives(base) -- do non-reflexive negative imperative slot overrides -- FIXME: What about reflexive negative imperatives? process_slot_overrides(base, function(slot)		return slot:find("^neg_imp_")	end) -- This should happen before add_missing_links_to_forms so that the comparison `form == base.lemma` -- in handle_infinitive_linked works correctly and compares unlinked forms to unlinked forms. handle_infinitive_linked(base) if not base.alternant_multiword_spec.args.noautolinkverb then add_missing_links_to_forms(base) end remove_mesoclitic_insertion_points(base) process_superseded_forms(base) end

local function parse_indicator_spec(angle_bracket_spec) -- Store the original angle bracket spec so we can reconstruct the overall conj spec with the lemma(s) in them. local base = { angle_bracket_spec = angle_bracket_spec, user_basic_overrides = {}, user_stems = {}, }	local function parse_err(msg) error(msg .. ": " .. angle_bracket_spec) end local function fetch_footnotes(separated_group) local footnotes for j = 2, #separated_group - 1, 2 do			if separated_group[j + 1] ~= "" then parse_err("Extraneous text after bracketed footnotes: '" .. table.concat(separated_group) .. "'") end if not footnotes then footnotes = {} end table.insert(footnotes, separated_group[j]) end return footnotes end

local inside = angle_bracket_spec:match("^<(.*)>$") assert(inside) if inside == "" then return base end local segments = iut.parse_balanced_segment_run(inside, "[", "]") local dot_separated_groups = iut.split_alternating_runs(segments, "%.") for i, dot_separated_group in ipairs(dot_separated_groups) do		if rfind(dot_separated_group[1], ":") then local colon_separated_groups = iut.split_alternating_runs(dot_separated_group, "%s*:%s*") local first_element = colon_separated_groups[1][1] if #colon_separated_groups[1] > 1 then parse_err("Can't attach footnotes directly to '" .. first_element .. "' spec; attach them to the " ..					"colon-separated values following the initial colon") end if overridable_stems[first_element] then if base.user_stems[first_element] then parse_err("Overridable stem '" .. first_element .. "' specified twice") end table.remove(colon_separated_groups, 1) base.user_stems[first_element] = overridable_stems[first_element](colon_separated_groups,					{prefix = first_element, base = base, parse_err = parse_err, fetch_footnotes = fetch_footnotes}) else -- assume a basic override; we validate further later when the possible slots are available if base.user_basic_overrides[first_element] then parse_err("Basic override '" .. first_element .. "' specified twice") end table.remove(colon_separated_groups, 1) base.user_basic_overrides[first_element] = allow_multiple_values(colon_separated_groups,					{prefix = first_element, base = base, parse_err = parse_err, fetch_footnotes = fetch_footnotes}) end else local comma_separated_groups = iut.split_alternating_runs(dot_separated_group, "%s*,%s*") local first_element = comma_separated_groups[1][1] if vowel_alternants[first_element] then for j = 1, #comma_separated_groups do					local alt = comma_separated_groups[j][1] if not vowel_alternants[alt] then parse_err("Unrecognized vowel alternant '" .. alt .. "'") end if base.vowel_alt then for _, existing_alt in ipairs(base.vowel_alt) do							if existing_alt.form == alt then parse_err("Vowel alternant '" .. alt .. "' specified twice") end end else base.vowel_alt = {} end table.insert(base.vowel_alt, {form = alt, footnotes = fetch_footnotes(comma_separated_groups[j])}) end elseif indicator_flags[first_element] then if #comma_separated_groups > 1 then parse_err("Multiple comma-separated elements not allowed with '" .. first_element .. "' spec") end if #comma_separated_groups[1] > 1 then parse_err("No footnotes allowed with '" .. first_element .. "' spec") end if base[first_element] then parse_err("Spec '" .. first_element .. "' specified twice") end base[first_element] = true else parse_err("Unrecognized spec '" .. comma_separated_groups[1][1] .. "'") end end end

return base end

-- Reconstruct the overall verb spec from the output of iut.parse_inflected_text, so we can use it in -- Module:accel/pt. function export.reconstruct_verb_spec(alternant_multiword_spec) local parts = {}

for _, alternant_or_word_spec in ipairs(alternant_multiword_spec.alternant_or_word_specs) do		table.insert(parts, alternant_or_word_spec.user_specified_before_text) if alternant_or_word_spec.alternants then table.insert(parts, "((") for i, multiword_spec in ipairs(alternant_or_word_spec.alternants) do				if i > 1 then table.insert(parts, ",") end for _, word_spec in ipairs(multiword_spec.word_specs) do					table.insert(parts, word_spec.user_specified_before_text) table.insert(parts, word_spec.user_specified_lemma) table.insert(parts, word_spec.angle_bracket_spec) end table.insert(parts, multiword_spec.user_specified_post_text) end table.insert(parts, "))") else table.insert(parts, alternant_or_word_spec.user_specified_lemma) table.insert(parts, alternant_or_word_spec.angle_bracket_spec) end end table.insert(parts, alternant_multiword_spec.user_specified_post_text)

-- As a special case, if we see e.g. "amar<>", remove the <>. Don't do this if there are spaces or alternants. local retval = table.concat(parts) if not retval:find(" ") and not retval:find("%(%(") then local retval_no_angle_brackets = retval:match("^(.*)<>$") if retval_no_angle_brackets then return retval_no_angle_brackets end end return retval end

-- Normalize all lemmas, substituting the pagename for blank lemmas and adding links to multiword lemmas. local function normalize_all_lemmas(alternant_multiword_spec, pagename)

-- (1) Add links to all before and after text. Remember the original text so we can reconstruct the verb spec later. if not alternant_multiword_spec.args.noautolinktext then for _, alternant_or_word_spec in ipairs(alternant_multiword_spec.alternant_or_word_specs) do			alternant_or_word_spec.user_specified_before_text = alternant_or_word_spec.before_text alternant_or_word_spec.before_text = com.add_links(alternant_or_word_spec.before_text) if alternant_or_word_spec.alternants then for _, multiword_spec in ipairs(alternant_or_word_spec.alternants) do					for _, word_spec in ipairs(multiword_spec.word_specs) do						word_spec.user_specified_before_text = word_spec.before_text word_spec.before_text = com.add_links(word_spec.before_text) end multiword_spec.user_specified_post_text = multiword_spec.post_text multiword_spec.post_text = com.add_links(multiword_spec.post_text) end end end alternant_multiword_spec.user_specified_post_text = alternant_multiword_spec.post_text alternant_multiword_spec.post_text = com.add_links(alternant_multiword_spec.post_text) end

-- (2) Remove any links from the lemma, but remember the original form --    so we can use it below in the 'lemma_linked' form. iut.map_word_specs(alternant_multiword_spec, function(base)		if base.lemma == "" then			base.lemma = pagename		end

base.user_specified_lemma = base.lemma

base.lemma = m_links.remove_links(base.lemma) local refl_verb = base.lemma local verb, refl = rmatch(refl_verb, "^(.-)%-(se)$") if not verb then verb, refl = refl_verb, nil end base.user_specified_verb = verb base.refl = refl base.verb = base.user_specified_verb

local linked_lemma if alternant_multiword_spec.args.noautolinkverb or base.user_specified_lemma:find("%[%[") then linked_lemma = base.user_specified_lemma elseif base.refl then -- Reconstruct the linked lemma with separate links around base verb and reflexive pronoun. linked_lemma = base.user_specified_verb == base.verb and "" .. base.user_specified_verb .. "" or "" .. base.user_specified_verb .. "" linked_lemma = linked_lemma .. (refl and "-" .. refl .. "" or "") else -- Add links to the lemma so the user doesn't specifically need to, since we preserve -- links in multiword lemmas and include links in non-lemma forms rather than allowing -- the entire form to be a link. linked_lemma = com.add_links(base.user_specified_lemma) end base.linked_lemma = linked_lemma end) end

local function detect_indicator_spec(base) if (base.only3s and 1 or 0) + (base.only3sp and 1 or 0) + (base.only3p and 1 or 0) > 1 then error("Only one of 'only3s', 'only3sp' and 'only3p' can be specified") end

base.forms = {} base.stems = {} base.basic_overrides = {} base.basic_reflexive_only_overrides = {} if not base.no_built_in then for _, built_in_conj in ipairs(built_in_conjugations) do			if type(built_in_conj.match) == "function" then base.prefix, base.non_prefixed_verb = built_in_conj.match(base.verb) elseif built_in_conj.match:find("^%^") and rsub(built_in_conj.match, "^%^", "") == base.verb then -- begins with ^, for exact match, and matches base.prefix, base.non_prefixed_verb = "", base.verb else base.prefix, base.non_prefixed_verb = rmatch(base.verb, "^(.*)(" .. built_in_conj.match .. ")$")			end if base.prefix then -- we found a built-in verb for stem, forms in pairs(built_in_conj.forms) do					if type(forms) == "function" then forms = forms(base, base.prefix) end if stem:find("^refl_") then stem = stem:gsub("^refl_", "") if not base.alternant_multiword_spec.verb_slots_basic_map[stem] then error("Internal error: setting for 'refl_" .. stem .. "' does not refer to a basic verb slot") end base.basic_reflexive_only_overrides[stem] = forms elseif base.alternant_multiword_spec.verb_slots_basic_map[stem] then -- an individual form override of a basic form base.basic_overrides[stem] = forms else base.stems[stem] = forms end end break end end end

-- Override built-in-verb stems and overrides with user-specified ones. for stem, values in pairs(base.user_stems) do		base.stems[stem] = values end for override, values in pairs(base.user_basic_overrides) do		if not base.alternant_multiword_spec.verb_slots_basic_map[override] then error("Unrecognized override '" .. override .. "': " .. base.angle_bracket_spec) end base.basic_overrides[override] = values end

base.prefix = base.prefix or "" base.non_prefixed_verb = base.non_prefixed_verb or base.verb local inf_stem, suffix = rmatch(base.non_prefixed_verb, "^(.*)([aeioô]r)$") if not inf_stem then error("Unrecognized infinitive: " .. base.verb) end base.inf_stem = inf_stem suffix = suffix == "ôr" and "or" or suffix base.conj = suffix base.conj_vowel = suffix == "ar" and "á" or suffix == "ir" and "í" or "ê" base.frontback = suffix == "ar" and "back" or "front"

if base.stems.vowel_alt then -- built-in verb with specified vowel alternation if base.vowel_alt then error(base.verb .. " is a recognized built-in verb, and should not have vowel alternations specified with it") end base.vowel_alt = iut.convert_to_general_list_form(base.stems.vowel_alt) end -- Propagate built-in-verb indicator flags to `base` and combine with user-specified flags. for indicator_flag, _ in pairs(indicator_flags) do		base[indicator_flag] = base[indicator_flag] or base.stems[indicator_flag] end

-- Convert vowel alternation indicators into stems. local vowel_alt = base.vowel_alt or 	base.vowel_alt_stems = apply_vowel_alternations(base.inf_stem, vowel_alt) for _, vowel_alt_stems in ipairs(base.vowel_alt_stems) do		if vowel_alt_stems.err then error("To use '" .. vowel_alt_stems.altobj.form .. "', present stem '" .. base.prefix .. base.inf_stem .. "' " ..				vowel_alt_stems.err) end end end

local function detect_all_indicator_specs(alternant_multiword_spec) -- Propagate some settings up; some are used internally, others by Module:pt-headword. iut.map_word_specs(alternant_multiword_spec, function(base)		-- Internal indicator flags. Do these before calling detect_indicator_spec because add_slots uses them.		for _, prop in ipairs { "refl", "clitic" } do			if base[prop] then				alternant_multiword_spec[prop] = true			end		end		base.alternant_multiword_spec = alternant_multiword_spec	end)

add_slots(alternant_multiword_spec)

alternant_multiword_spec.vowel_alt = {} iut.map_word_specs(alternant_multiword_spec, function(base)		detect_indicator_spec(base)		-- User-specified indicator flags. Do these after calling detect_indicator_spec because the latter may set these		-- indicators for built-in verbs.		for prop, _ in pairs(indicator_flags) do			if base[prop] then				alternant_multiword_spec[prop] = true			end		end		-- Vowel alternants. Do these after calling detect_indicator_spec because the latter sets base.vowel_alt for		-- built-in verbs.		if base.vowel_alt then			for _, altobj in ipairs(base.vowel_alt) do				m_table.insertIfNot(alternant_multiword_spec.vowel_alt, altobj.form)			end		end	end) end

local function add_categories_and_annotation(alternant_multiword_spec, base, multiword_lemma) local function insert_ann(anntype, value) m_table.insertIfNot(alternant_multiword_spec.annotation[anntype], value) end

local function insert_cat(cat, also_when_multiword) -- Don't place multiword terms in categories like 'Portuguese verbs ending in -ar' to avoid spamming the -- categories with such terms. if also_when_multiword or not multiword_lemma then m_table.insertIfNot(alternant_multiword_spec.categories, "Portuguese " .. cat) end end

if check_for_red_links and not alternant_multiword_spec.from_headword and not alternant_multiword_spec.from_verb_form_of and multiword_lemma then for _, slot_and_accel in ipairs(alternant_multiword_spec.all_verb_slots) do			local slot = slot_and_accel[1] local forms = base.forms[slot] local must_break = false if forms then for _, form in ipairs(forms) do					if not form.form:find("%[%[") then local title = mw.title.new(form.form) if title and not title.exists then insert_cat("verbs with red links in their inflection tables") must_break = true break end end end end if must_break then break end end end

insert_cat("verbs ending in -" .. base.conj)

if base.irreg then insert_ann("irreg", "irregular") insert_cat("irregular verbs") else insert_ann("irreg", "regular") end

if base.only3s then insert_ann("defective", "impersonal") insert_cat("impersonal verbs") elseif base.only3sp then insert_ann("defective", "third-person only") insert_cat("third-person-only verbs") elseif base.only3p then insert_ann("defective", "third-person plural only") insert_cat("third-person-plural-only verbs") elseif base.no_pres_stressed or base.no_pres1_and_sub then insert_ann("defective", "defective") insert_cat("defective verbs") else insert_ann("defective", "regular") end

if base.stems.short_pp then insert_ann("short_pp", "irregular short past participle") insert_cat("verbs with irregular short past participle") else insert_ann("short_pp", "regular") end

if base.clitic then insert_cat("verbs with lexical clitics") end

if base.refl then insert_cat("reflexive verbs") end

if base.e_ei_cat then insert_ann("vowel_alt", "e becomes ei when stressed") insert_cat("verbs with e becoming ei when stressed") elseif not base.vowel_alt then insert_ann("vowel_alt", "non-alternating") else for _, alt in ipairs(base.vowel_alt) do			if alt.form == "+" then insert_ann("vowel_alt", "non-alternating") else insert_ann("vowel_alt", vowel_alternant_to_desc[alt.form]) insert_cat("verbs with " .. vowel_alternant_to_cat[alt.form]) end end end

local cons_alt = base.stems.cons_alt if cons_alt == nil then if base.conj == "ar" then if base.inf_stem:find("ç$") then cons_alt = "c-ç" elseif base.inf_stem:find("c$") then cons_alt = "c-qu" elseif base.inf_stem:find("g$") then cons_alt = "g-gu" end else if base.no_pres_stressed or base.no_pres1_and_sub then cons_alt = nil -- no e.g. c-ç alternation in this case elseif base.inf_stem:find("c$") then cons_alt = "c-ç" elseif base.inf_stem:find("qu$") then cons_alt = "c-qu" elseif base.inf_stem:find("g$") then cons_alt = "g-j" elseif base.inf_stem:find("gu$") then cons_alt = "g-gu" end end end

if cons_alt then local desc = cons_alt .. " alternation" insert_ann("cons_alt", desc) insert_cat("verbs with " .. desc) else insert_ann("cons_alt", "non-alternating") end end

-- Compute the categories to add the verb to, as well as the annotation to display in the -- conjugation title bar. We combine the code to do these functions as both categories and -- title bar contain similar information. local function compute_categories_and_annotation(alternant_multiword_spec) alternant_multiword_spec.categories = {} local ann = {} alternant_multiword_spec.annotation = ann ann.irreg = {} ann.short_pp = {} ann.defective = {} ann.vowel_alt = {} ann.cons_alt = {}

local multiword_lemma = false for _, form in ipairs(alternant_multiword_spec.forms.infinitive) do		if form.form:find(" ") then multiword_lemma = true break end end

iut.map_word_specs(alternant_multiword_spec, function(base)		add_categories_and_annotation(alternant_multiword_spec, base, multiword_lemma)	end) local ann_parts = {} local irreg = table.concat(ann.irreg, " or ") if irreg ~= "" and irreg ~= "regular" then table.insert(ann_parts, irreg) end local short_pp = table.concat(ann.short_pp, " or ") if short_pp ~= "" and short_pp ~= "regular" then table.insert(ann_parts, short_pp) end local defective = table.concat(ann.defective, " or ") if defective ~= "" and defective ~= "regular" then table.insert(ann_parts, defective) end local vowel_alt = table.concat(ann.vowel_alt, " or ") if vowel_alt ~= "" and vowel_alt ~= "non-alternating" then table.insert(ann_parts, vowel_alt) end local cons_alt = table.concat(ann.cons_alt, " or ") if cons_alt ~= "" and cons_alt ~= "non-alternating" then table.insert(ann_parts, cons_alt) end alternant_multiword_spec.annotation = table.concat(ann_parts, "; ") end

local function show_forms(alternant_multiword_spec) local lemmas = alternant_multiword_spec.forms.infinitive alternant_multiword_spec.lemmas = lemmas -- save for later use in make_table

if alternant_multiword_spec.forms.short_pp_ms then alternant_multiword_spec.has_short_pp = true end local reconstructed_verb_spec = export.reconstruct_verb_spec(alternant_multiword_spec)

local function transform_accel_obj(slot, formobj, accel_obj) -- No accelerators for negative imperatives, which are always multiword and derived directly from the -- present subjunctive. if slot:find("^neg_imp") then return nil end if accel_obj then accel_obj.form = "verb-form-" .. reconstructed_verb_spec end return accel_obj end

-- Italicize superseded forms. local function generate_link(slot, form, origentry, accel_obj) if origentry:find(VAR_SUPERSEDED) then origentry = rsub(origentry, VAR_SUPERSEDED, "") return m_links.full_link({lang = lang, term = origentry, tr = "-", accel = accel_obj}, "term") end end

local props = { lang = lang, lemmas = lemmas, transform_accel_obj = transform_accel_obj, canonicalize = function(form) return export.remove_variant_codes(form, "keep superseded") end, generate_link = generate_link, slot_list = alternant_multiword_spec.verb_slots_basic, }	iut.show_forms(alternant_multiword_spec.forms, props) alternant_multiword_spec.footnote_basic = alternant_multiword_spec.forms.footnote end

local notes_template = [=[ {footnote} ]=]

local basic_table = [=[ {description} Conjugation of {title} Conjugation of {title} (See Appendix:Portuguese verbs) {\op}| class="inflection-table" style="background:#F6F6F6; text-align: left; border: 1px solid #999999;" cellpadding="3" cellspacing="0" ! style="border: 1px solid #999999; background:#B0B0B0" rowspan="2" | ! style="border: 1px solid #999999; background:#D0D0D0" colspan="3" | Singular ! style="border: 1px solid #999999; background:#D0D0D0" colspan="3" | Plural ! style="border: 1px solid #999999; background:#D0D0D0; width:12.5%" | First-person (<<eu>>) ! style="border: 1px solid #999999; background:#D0D0D0; width:12.5%" | Second-person (<<tu>>) ! style="border: 1px solid #999999; background:#D0D0D0; width:12.5%" | Third-person (< > / < > / <<você>>) ! style="border: 1px solid #999999; background:#D0D0D0; width:12.5%" | First-person (<<nós>>) ! style="border: 1px solid #999999; background:#D0D0D0; width:12.5%" | Second-person (<<vós>>) ! style="border: 1px solid #999999; background:#D0D0D0; width:12.5%" | Third-person (< > / < > / <<vocês>>) ! style="border: 1px solid #999999; background:#c498ff" colspan="7" |  Infinitive  ! style="border: 1px solid #999999; background:#a478df" |  Impersonal  ! style="border: 1px solid #999999; background:#a478df" |  Personal  ! style="border: 1px solid #999999; background:#98ffc4" colspan="7" | <span title="gerúndio">Gerund  ! style="border: 1px solid #999999; background:#d0dff4" colspan="7" |  Indicative  ! style="border: 1px solid #999999; background:#b0bfd4" | Present ! style="border: 1px solid #999999; background:#b0bfd4" | <span title="pretérito imperfeito">Imperfect ! style="border: 1px solid #999999; background:#b0bfd4" | <span title="pretérito perfeito">Preterite ! style="border: 1px solid #999999; background:#b0bfd4" | <span title="pretérito mais-que-perfeito simples">Pluperfect ! style="border: 1px solid #999999; background:#b0bfd4" | Future ! style="border: 1px solid #999999; background:#ffffaa" colspan="7" | <span title="condicional / futuro do pretérito">Conditional  ! style="border: 1px solid #999999; background:#ddddaa" | ! style="border: 1px solid #999999; background:#d0f4d0" colspan="7" | <span title="conjuntivo (pt) / subjuntivo (br)">Subjunctive  ! style="border: 1px solid #999999; background:#b0d4b0" | <span title=" presente do conjuntivo (pt) / subjuntivo (br)">Present ! style="border: 1px solid #999999; background:#b0d4b0" | <span title="pretérito imperfeito do conjuntivo (pt) / subjuntivo (br)">Imperfect ! style="border: 1px solid #999999; background:#b0d4b0" | <span title="futuro do conjuntivo (pt) / subjuntivo (br)">Future ! style="border: 1px solid #999999; background:#f4e4d0" colspan="7" |  Imperative  ! style="border: 1px solid #999999; background:#d4c4b0" | Affirmative ! style="border: 1px solid #999999; background:#d4c4b0" | Negative (<<não>>) ]=]
 * style="border: 1px solid #999999; vertical-align: top;" colspan="6" | {infinitive}
 * style="border: 1px solid #999999; vertical-align: top;" | {pers_inf_1s}
 * style="border: 1px solid #999999; vertical-align: top;" | {pers_inf_2s}
 * style="border: 1px solid #999999; vertical-align: top;" | {pers_inf_3s}
 * style="border: 1px solid #999999; vertical-align: top;" | {pers_inf_1p}
 * style="border: 1px solid #999999; vertical-align: top;" | {pers_inf_2p}
 * style="border: 1px solid #999999; vertical-align: top;" | {pers_inf_3p}
 * style="border: 1px solid #999999; background:#78dfa4" |
 * style="border: 1px solid #999999; vertical-align: top;" colspan="6" | {gerund}
 * -{pp_clause}
 * -{pp_clause}
 * style="border: 1px solid #999999; vertical-align: top;" | {pres_1s}
 * style="border: 1px solid #999999; vertical-align: top;" | {pres_2s}
 * style="border: 1px solid #999999; vertical-align: top;" | {pres_3s}
 * style="border: 1px solid #999999; vertical-align: top;" | {pres_1p}
 * style="border: 1px solid #999999; vertical-align: top;" | {pres_2p}
 * style="border: 1px solid #999999; vertical-align: top;" | {pres_3p}
 * style="border: 1px solid #999999; vertical-align: top;" | {impf_1s}
 * style="border: 1px solid #999999; vertical-align: top;" | {impf_2s}
 * style="border: 1px solid #999999; vertical-align: top;" | {impf_3s}
 * style="border: 1px solid #999999; vertical-align: top;" | {impf_1p}
 * style="border: 1px solid #999999; vertical-align: top;" | {impf_2p}
 * style="border: 1px solid #999999; vertical-align: top;" | {impf_3p}
 * style="border: 1px solid #999999; vertical-align: top;" | {pret_1s}
 * style="border: 1px solid #999999; vertical-align: top;" | {pret_2s}
 * style="border: 1px solid #999999; vertical-align: top;" | {pret_3s}
 * style="border: 1px solid #999999; vertical-align: top;" | {pret_1p}
 * style="border: 1px solid #999999; vertical-align: top;" | {pret_2p}
 * style="border: 1px solid #999999; vertical-align: top;" | {pret_3p}
 * style="border: 1px solid #999999; vertical-align: top;" | {plup_1s}
 * style="border: 1px solid #999999; vertical-align: top;" | {plup_2s}
 * style="border: 1px solid #999999; vertical-align: top;" | {plup_3s}
 * style="border: 1px solid #999999; vertical-align: top;" | {plup_1p}
 * style="border: 1px solid #999999; vertical-align: top;" | {plup_2p}
 * style="border: 1px solid #999999; vertical-align: top;" | {plup_3p}
 * style="border: 1px solid #999999; vertical-align: top;" | {fut_1s}
 * style="border: 1px solid #999999; vertical-align: top;" | {fut_2s}
 * style="border: 1px solid #999999; vertical-align: top;" | {fut_3s}
 * style="border: 1px solid #999999; vertical-align: top;" | {fut_1p}
 * style="border: 1px solid #999999; vertical-align: top;" | {fut_2p}
 * style="border: 1px solid #999999; vertical-align: top;" | {fut_3p}
 * style="border: 1px solid #999999; vertical-align: top;" | {cond_1s}
 * style="border: 1px solid #999999; vertical-align: top;" | {cond_2s}
 * style="border: 1px solid #999999; vertical-align: top;" | {cond_3s}
 * style="border: 1px solid #999999; vertical-align: top;" | {cond_1p}
 * style="border: 1px solid #999999; vertical-align: top;" | {cond_2p}
 * style="border: 1px solid #999999; vertical-align: top;" | {cond_3p}
 * style="border: 1px solid #999999; vertical-align: top;" | {pres_sub_1s}
 * style="border: 1px solid #999999; vertical-align: top;" | {pres_sub_2s}
 * style="border: 1px solid #999999; vertical-align: top;" | {pres_sub_3s}
 * style="border: 1px solid #999999; vertical-align: top;" | {pres_sub_1p}
 * style="border: 1px solid #999999; vertical-align: top;" | {pres_sub_2p}
 * style="border: 1px solid #999999; vertical-align: top;" | {pres_sub_3p}
 * style="border: 1px solid #999999; vertical-align: top;" | {impf_sub_1s}
 * style="border: 1px solid #999999; vertical-align: top;" | {impf_sub_2s}
 * style="border: 1px solid #999999; vertical-align: top;" | {impf_sub_3s}
 * style="border: 1px solid #999999; vertical-align: top;" | {impf_sub_1p}
 * style="border: 1px solid #999999; vertical-align: top;" | {impf_sub_2p}
 * style="border: 1px solid #999999; vertical-align: top;" | {impf_sub_3p}
 * style="border: 1px solid #999999; vertical-align: top;" | {fut_sub_1s}
 * style="border: 1px solid #999999; vertical-align: top;" | {fut_sub_2s}
 * style="border: 1px solid #999999; vertical-align: top;" | {fut_sub_3s}
 * style="border: 1px solid #999999; vertical-align: top;" | {fut_sub_1p}
 * style="border: 1px solid #999999; vertical-align: top;" | {fut_sub_2p}
 * style="border: 1px solid #999999; vertical-align: top;" | {fut_sub_3p}
 * style="border: 1px solid #999999; vertical-align: top;" rowspan="2" |
 * style="border: 1px solid #999999; vertical-align: top;" | {imp_2s}
 * style="border: 1px solid #999999; vertical-align: top;" | {imp_3s}
 * style="border: 1px solid #999999; vertical-align: top;" | {imp_1p}
 * style="border: 1px solid #999999; vertical-align: top;" | {imp_2p}
 * style="border: 1px solid #999999; vertical-align: top;" | {imp_3p}
 * style="border: 1px solid #999999; vertical-align: top;" | {neg_imp_2s}
 * style="border: 1px solid #999999; vertical-align: top;" | {neg_imp_3s}
 * style="border: 1px solid #999999; vertical-align: top;" | {neg_imp_1p}
 * style="border: 1px solid #999999; vertical-align: top;" | {neg_imp_2p}
 * style="border: 1px solid #999999; vertical-align: top;" | {neg_imp_3p}
 * {\cl}{notes_clause}

local double_pp_template = [=[

! style="border: 1px solid #999999; background:#ffc498" colspan="7" | <span title="particípio irregular">Short past participle  ! style="border: 1px solid #999999; background:#dfa478" | Masculine ! style="border: 1px solid #999999; background:#dfa478" | Feminine ! style="border: 1px solid #999999; background:#ffc498" colspan="7" | <span title="particípio regular">Long past participle  ! style="border: 1px solid #999999; background:#dfa478" | Masculine ! style="border: 1px solid #999999; background:#dfa478" | Feminine
 * style="border: 1px solid #999999; vertical-align: top;" colspan="3" | {short_pp_ms}
 * style="border: 1px solid #999999; vertical-align: top;" colspan="3" | {short_pp_mp}
 * style="border: 1px solid #999999; vertical-align: top;" colspan="3" | {short_pp_fs}
 * style="border: 1px solid #999999; vertical-align: top;" colspan="3" | {short_pp_fp}
 * style="border: 1px solid #999999; vertical-align: top;" colspan="3" | {pp_ms}
 * style="border: 1px solid #999999; vertical-align: top;" colspan="3" | {pp_mp}
 * style="border: 1px solid #999999; vertical-align: top;" colspan="3" | {pp_fs}
 * style="border: 1px solid #999999; vertical-align: top;" colspan="3" | {pp_fp}
 * -]=]

local single_pp_template = [=[

! style="border: 1px solid #999999; background:#ffc498" colspan="7" | <span title="particípio passado">Past participle  ! style="border: 1px solid #999999; background:#dfa478" | Masculine ! style="border: 1px solid #999999; background:#dfa478" | Feminine
 * style="border: 1px solid #999999; vertical-align: top;" colspan="3" | {pp_ms}
 * style="border: 1px solid #999999; vertical-align: top;" colspan="3" | {pp_mp}
 * style="border: 1px solid #999999; vertical-align: top;" colspan="3" | {pp_fs}
 * style="border: 1px solid #999999; vertical-align: top;" colspan="3" | {pp_fp}
 * -]=]

local function make_table(alternant_multiword_spec) local forms = alternant_multiword_spec.forms

forms.title = link_term(alternant_multiword_spec.lemmas[1].form) if alternant_multiword_spec.annotation ~= "" then forms.title = forms.title .. " (" .. alternant_multiword_spec.annotation .. ")" end forms.description = ""

-- Format the table. forms.footnote = alternant_multiword_spec.footnote_basic forms.notes_clause = forms.footnote ~= "" and m_string_utilities.format(notes_template, forms) or "" -- has_short_pp is computed in show_forms. local pp_template = alternant_multiword_spec.has_short_pp and double_pp_template or single_pp_template forms.pp_clause = m_string_utilities.format(pp_template, forms) local table_with_pronouns = rsub(basic_table, "<<(.-)>>", link_term) return m_string_utilities.format(table_with_pronouns, forms) end

-- Externally callable function to parse and conjugate a verb given user-specified arguments. -- Return value is WORD_SPEC, an object where the conjugated forms are in `WORD_SPEC.forms` -- for each slot. If there are no values for a slot, the slot key will be missing. The value -- for a given slot is a list of objects {form=FORM, footnotes=FOOTNOTES}. function export.do_generate_forms(parent_args, from_headword, from_verb_form_of) local params = { [1] = {required = from_verb_form_of}, ["noautolinktext"] = {type = "boolean"}, ["noautolinkverb"] = {type = "boolean"}, ["pagename"] = {}, -- for testing/documentation pages ["json"] = {type = "boolean"}, -- for bot use }

if from_headword then params["head"] = {list = true} params["pres"] = {list = true} --present params["pres_qual"] = {list = "pres=_qual", allow_holes = true} params["pret"] = {list = true} --preterite params["pret_qual"] = {list = "pret=_qual", allow_holes = true} params["part"] = {list = true} --participle params["part_qual"] = {list = "part=_qual", allow_holes = true} params["short_part"] = {list = true} --short participle params["short_part_qual"] = {list = "part=_qual", allow_holes = true} params["attn"] = {type = "boolean"} params["id"] = {} end

local args = require("Module:parameters").process(parent_args, params) local PAGENAME = mw.title.getCurrentTitle.text local function in_template_space return mw.title.getCurrentTitle.nsText == "Template" end

-- Determine the verb spec we're being asked to generate the conjugation of. This may be taken from the -- current page title or the value of |pagename=; but not when called from, where the -- page title is a non-lemma form. Note that the verb spec may omit the infinitive; e.g. it may be "<i-e>". -- For this reason, we use the value of `pagename` computed here down below, when calling normalize_all_lemmas. local pagename = not from_verb_form_of and args.pagename or from_headword and args.head[1] or PAGENAME local arg1 = args[1] if not arg1 then if (pagename == "pt-conj" or pagename == "pt-verb") and in_template_space then arg1 = "cergir<i-e,i>" elseif pagename == "pt-verb form of" and in_template_space then arg1 = "amar" else arg1 = pagename end end

-- When called from, determine the non-lemma form whose inflections we're being asked to -- determine. This normally comes from the page title or the value of |pagename=. local verb_form_of_form if from_verb_form_of then verb_form_of_form = args.pagename if not verb_form_of_form then if PAGENAME == "pt-verb form of" and in_template_space then verb_form_of_form = "ame" else verb_form_of_form = PAGENAME end end end

if arg1:find(" ") and not arg1:find("<") then -- If multiword lemma without <> already, try to add it after the first word.

local need_explicit_angle_brackets = false if arg1:find("%(%(") then need_explicit_angle_brackets = true else local refl_clitic_verb, orig_refl_clitic_verb, post

-- Try to preserve the brackets in the part after the verb, but don't do it			-- if there aren't the same number of left and right brackets in the verb -- (which means the verb was linked as part of a larger expression). refl_clitic_verb, post = rmatch(arg1, "^(.-)( .*)$") local left_brackets = rsub(refl_clitic_verb, "[^%[]", "") local right_brackets = rsub(refl_clitic_verb, "[^%]]", "") if #left_brackets == #right_brackets then arg1 = refl_clitic_verb .. "<>" .. post else need_explicit_angle_brackets = true end end

if need_explicit_angle_brackets then error("Multiword argument without <> and with alternants, a multiword linked verb or unbalanced brackets; please include <> explicitly: " .. arg1) end end

local function split_bracketed_runs_into_words(bracketed_runs) return iut.split_alternating_runs(bracketed_runs, " ", "preserve splitchar") end

local parse_props = { parse_indicator_spec = parse_indicator_spec, -- Split words only on spaces, not on hyphens, because that messes up reflexive verb parsing. split_bracketed_runs_into_words = split_bracketed_runs_into_words, allow_default_indicator = true, allow_blank_lemma = true, }	local alternant_multiword_spec = iut.parse_inflected_text(arg1, parse_props) alternant_multiword_spec.pos = pos or "verbs" alternant_multiword_spec.args = args alternant_multiword_spec.from_headword = from_headword alternant_multiword_spec.from_verb_form_of = from_verb_form_of alternant_multiword_spec.verb_form_of_form = verb_form_of_form

normalize_all_lemmas(alternant_multiword_spec, pagename) detect_all_indicator_specs(alternant_multiword_spec) local inflect_props = { slot_list = alternant_multiword_spec.all_verb_slots, inflect_word_spec = conjugate_verb, get_variants = function(form) return rsub(form, not_var_code_c, "") end, -- We add links around the generated verbal forms rather than allow the entire multiword -- expression to be a link, so ensure that user-specified links get included as well. include_user_specified_links = true, }	iut.inflect_multiword_or_alternant_multiword_spec(alternant_multiword_spec, inflect_props)

-- Remove redundant brackets around entire forms. for slot, forms in pairs(alternant_multiword_spec.forms) do		for _, form in ipairs(forms) do			form.form = com.strip_redundant_links(form.form) end end

compute_categories_and_annotation(alternant_multiword_spec) if args.json and not from_headword and not from_verb_form_of then return export.remove_variant_codes(require("Module:JSON").toJSON(alternant_multiword_spec.forms)) end return alternant_multiword_spec end

-- Entry point for. Template-callable function to parse and conjugate a verb given -- user-specified arguments and generate a displayable table of the conjugated forms. function export.show(frame) local parent_args = frame:getParent.args local alternant_multiword_spec = export.do_generate_forms(parent_args) if type(alternant_multiword_spec) == "string" then -- JSON return value return alternant_multiword_spec end show_forms(alternant_multiword_spec) return make_table(alternant_multiword_spec) .. require("Module:utilities").format_categories(alternant_multiword_spec.categories, lang, nil, nil, force_cat) end

return export