(* Title: Pure/Syntax/parser.ML
Author: Carsten Clasohm, Sonia Mahjoub
Author: Makarius
General context-free parser for the inner syntax of terms and types.
*)
signature PARSER =
sig
type gram
val empty_gram: gram
val extend_gram: Syntax_Ext.xprod list -> gram -> gram
val make_gram: Syntax_Ext.xprod list -> gram
val pretty_gram: gram -> Pretty.T list
datatype parsetree =
Node of string * parsetree list |
Tip of Lexicon.token
exception PARSETREE of parsetree
val pretty_parsetree: parsetree -> Pretty.T
val parse: gram -> string -> Lexicon.token list -> parsetree list
val guess_infix_lr: gram -> string -> (string * bool * bool * int) option
val branching_level: int Config.T
end;
structure Parser: PARSER =
struct
(** datatype gram **)
(* nonterminals *)
(*production for the NTs are stored in a vector, indexed by the NT tag*)
type nt = int;
type tags = nt Symtab.table;
val tags_empty: tags = Symtab.empty;
fun tags_content (tags: tags) = sort_by #1 (Symtab.dest tags);
fun tags_lookup (tags: tags) = Symtab.lookup tags;
fun tags_insert tag (tags: tags) = Symtab.update_new tag tags;
fun tags_name (tags: tags) = the o Inttab.lookup (Inttab.make (map swap (Symtab.dest tags)));
type nts = Inttab.set;
val nts_empty: nts = Inttab.empty;
val nts_merge: nts * nts -> nts = Inttab.merge (K true);
fun nts_insert nt : nts -> nts = Inttab.insert_set nt;
fun nts_member (nts: nts) = Inttab.defined nts;
fun nts_fold f (nts: nts) = Inttab.fold (f o #1) nts;
fun nts_subset (nts1: nts, nts2: nts) = Inttab.forall (nts_member nts2 o #1) nts1;
fun nts_is_empty (nts: nts) = Inttab.is_empty nts;
fun nts_is_unique (nts: nts) = nts_is_empty nts orelse Inttab.is_single nts;
(* tokens *)
structure Tokens = Table(type key = Lexicon.token val ord = Lexicon.tokens_match_ord);
type tokens = Tokens.set;
val tokens_empty: tokens = Tokens.empty;
val tokens_merge: tokens * tokens -> tokens = Tokens.merge (K true);
fun tokens_insert tk : tokens -> tokens = Tokens.insert_set tk;
fun tokens_remove tk : tokens -> tokens = Tokens.remove_set tk;
fun tokens_member (tokens: tokens) = Tokens.defined tokens;
fun tokens_is_empty (tokens: tokens) = Tokens.is_empty tokens;
fun tokens_fold f (tokens: tokens) = Tokens.fold (f o #1) tokens;
val tokens_union = tokens_fold tokens_insert;
val tokens_subtract = tokens_fold tokens_remove;
fun tokens_find P (tokens: tokens) =
Tokens.get_first (fn (tok, ()) => if P tok then SOME tok else NONE) tokens;
fun tokens_add (nt: nt, tokens) = if tokens_is_empty tokens then I else cons (nt, tokens);
(* productions *)
datatype symb =
Terminal of Lexicon.token |
Nonterminal of nt * int; (*(tag, prio)*)
type prods = (symb list * string * int) list Tokens.table; (*start_token ~> [(rhs, name, prio)]*)
val prods_empty: prods = Tokens.empty;
fun prods_lookup (prods: prods) = Tokens.lookup_list prods;
fun prods_update entry : prods -> prods = Tokens.update entry;
fun prods_content (prods: prods) = distinct (op =) (maps #2 (Tokens.dest prods));
type nt_gram = (nts * tokens) * prods; (*dependent_nts, start_tokens, prods*)
(*depent_nts is a set of all NTs whose lookahead depends on this NT's lookahead*)
val nt_gram_empty: nt_gram = ((nts_empty, tokens_empty), prods_empty);
type chains = unit Int_Graph.T;
fun chains_preds (chains: chains) = Int_Graph.immediate_preds chains;
fun chains_all_preds (chains: chains) = Int_Graph.all_preds chains;
fun chains_all_succs (chains: chains) = Int_Graph.all_succs chains;
val chains_empty: chains = Int_Graph.empty;
fun chains_member (chains: chains) = Int_Graph.is_edge chains;
fun chains_declare nt : chains -> chains = Int_Graph.default_node (nt, ());
fun chains_insert (from, to) =
chains_declare from #> chains_declare to #> Int_Graph.add_edge (from, to);
datatype gram =
Gram of
{nt_count: int,
prod_count: int,
tags: tags,
chains: chains,
lambdas: nts,
prods: nt_gram Vector.vector};
(*"tags" is used to map NT names (i.e. strings) to tags;
chain productions are not stored as normal productions
but instead as an entry in "chains": from -> to;
lambda productions are stored as normal productions
and also as an entry in "lambdas"*)
(*productions for which no starting token is
known yet are associated with this token*)
val unknown_start = Lexicon.eof;
fun get_start tks =
(case Tokens.min tks of
SOME (tk, _) => tk
| NONE => unknown_start);
fun add_production prods (lhs, new_prod as (rhs, _, pri)) (chains, lambdas) =
let
(*store chain if it does not already exist*)
val (chain, new_chain, chains') =
(case (pri, rhs) of
(~1, [Nonterminal (from, ~1)]) =>
if chains_member chains (from, lhs)
then (SOME from, false, chains)
else (SOME from, true, chains_insert (from, lhs) chains)
| _ => (NONE, false, chains_declare lhs chains));
(*propagate new chain in lookahead and lambdas;
added_starts is used later to associate existing
productions with new starting tokens*)
val (added_starts, lambdas') =
if not new_chain then ([], lambdas)
else
let (*lookahead of chain's source*)
val ((_, from_tks), _) = Array.sub (prods, the chain);
(*copy from's lookahead to chain's destinations*)
fun copy_lookahead to =
let
val ((to_nts, to_tks), ps) = Array.sub (prods, to);
val new_tks = tokens_subtract to_tks from_tks; (*added lookahead tokens*)
val to_tks' = tokens_merge (to_tks, new_tks);
val _ = Array.update (prods, to, ((to_nts, to_tks'), ps));
in tokens_add (to, new_tks) end;
val tos = chains_all_succs chains' [lhs];
in
(fold copy_lookahead tos [],
lambdas |> nts_member lambdas lhs ? fold nts_insert tos)
end;
(*test if new production can produce lambda
(rhs must either be empty or only consist of lambda NTs)*)
val new_lambdas =
if forall
(fn Nonterminal (id, _) => nts_member lambdas' id
| Terminal _ => false) rhs
then SOME (filter_out (nts_member lambdas') (chains_all_succs chains' [lhs]))
else NONE;
val lambdas'' = fold nts_insert (these new_lambdas) lambdas';
(*list optional terminal and all nonterminals on which the lookahead
of a production depends*)
fun lookahead_dependency _ [] nts = (NONE, nts)
| lookahead_dependency _ (Terminal tk :: _) nts = (SOME tk, nts)
| lookahead_dependency lambdas (Nonterminal (nt, _) :: symbs) nts =
if nts_member lambdas nt then
lookahead_dependency lambdas symbs (nts_insert nt nts)
else (NONE, nts_insert nt nts);
(*get all known starting tokens for a nonterminal*)
fun starts_for_nt nt = snd (fst (Array.sub (prods, nt)));
(*update prods, lookaheads, and lambdas according to new lambda NTs*)
val (added_starts', lambdas') =
let
(*propagate added lambda NT*)
fun propagate_lambda [] added_starts lambdas = (added_starts, lambdas)
| propagate_lambda (l :: ls) added_starts lambdas =
let
(*get lookahead for lambda NT*)
val ((dependent, l_starts), _) = Array.sub (prods, l);
(*check productions whose lookahead may depend on lambda NT*)
fun examine_prods [] add_lambda nt_dependencies added_tks nt_prods =
(add_lambda, nt_dependencies, added_tks, nt_prods)
| examine_prods ((p as (rhs, _, _)) :: ps) add_lambda
nt_dependencies added_tks nt_prods =
let val (tk, nts) = lookahead_dependency lambdas rhs nts_empty in
if nts_member nts l then (*update production's lookahead*)
let
val new_lambda =
is_none tk andalso nts_subset (nts, lambdas);
val new_tks =
tokens_empty
|> fold tokens_insert (the_list tk)
|> nts_fold (tokens_union o starts_for_nt) nts
|> tokens_subtract l_starts;
val added_tks' = tokens_merge (added_tks, new_tks);
val nt_dependencies' = nts_merge (nt_dependencies, nts);
(*associate production with new starting tokens*)
fun copy tk nt_prods =
prods_update (tk, p :: prods_lookup nt_prods tk) nt_prods;
val nt_prods' = nt_prods
|> tokens_fold copy new_tks
|> new_lambda ? copy Lexicon.dummy;
in
examine_prods ps (add_lambda orelse new_lambda)
nt_dependencies' added_tks' nt_prods'
end
else (*skip production*)
examine_prods ps add_lambda nt_dependencies added_tks nt_prods
end;
(*check each NT whose lookahead depends on new lambda NT*)
fun process_nts nt (added_lambdas, added_starts) =
let
val ((old_nts, old_tks), nt_prods) = Array.sub (prods, nt);
(*existing productions whose lookahead may depend on l*)
val tk_prods = prods_lookup nt_prods (get_start l_starts);
(*add_lambda is true if an existing production of the nt
produces lambda due to the new lambda NT l*)
val (add_lambda, nt_dependencies, added_tks, nt_prods') =
examine_prods tk_prods false nts_empty tokens_empty nt_prods;
val new_nts = nts_merge (old_nts, nt_dependencies);
val new_tks = tokens_merge (old_tks, added_tks);
val added_lambdas' = added_lambdas |> add_lambda ? cons nt;
val _ = Array.update (prods, nt, ((new_nts, new_tks), nt_prods'));
(*N.B. that because the tks component
is used to access existing
productions we have to add new
tokens at the _end_ of the list*)
val added_starts' = tokens_add (nt, added_tks) added_starts;
in (added_lambdas', added_starts') end;
val (added_lambdas, added_starts') =
nts_fold process_nts dependent ([], added_starts);
val added_lambdas' = filter_out (nts_member lambdas) added_lambdas;
in
propagate_lambda (ls @ added_lambdas') added_starts'
(fold nts_insert added_lambdas' lambdas)
end;
in
propagate_lambda
(nts_fold (fn l => not (nts_member lambdas l) ? cons l) lambdas'' [])
added_starts lambdas''
end;
(*insert production into grammar*)
val added_starts' =
if is_some chain then added_starts' (*don't store chain production*)
else
let
(*lookahead tokens of new production and on which
NTs lookahead depends*)
val (start_tk, start_nts) = lookahead_dependency lambdas' rhs nts_empty;
val start_tks =
tokens_empty
|> fold tokens_insert (the_list start_tk)
|> nts_fold (tokens_union o starts_for_nt) start_nts;
val start_tks' =
start_tks
|> (if is_some new_lambdas then tokens_insert Lexicon.dummy
else if tokens_is_empty start_tks then tokens_insert unknown_start
else I);
(*add lhs NT to list of dependent NTs in lookahead*)
fun add_nts nt initial =
(if initial then
let val ((old_nts, old_tks), ps) = Array.sub (prods, nt) in
if nts_member old_nts lhs then ()
else Array.update (prods, nt, ((nts_insert lhs old_nts, old_tks), ps))
end
else (); false);
(*add new start tokens to chained NTs' lookahead list;
also store new production for lhs NT*)
fun add_tks [] added = added
| add_tks (nt :: nts) added =
let
val ((old_nts, old_tks), nt_prods) = Array.sub (prods, nt);
val new_tks = tokens_subtract old_tks start_tks;
(*store new production*)
fun store tk (prods, _) =
let
val tk_prods = prods_lookup prods tk;
val tk_prods' = new_prod :: tk_prods;
val prods' = prods_update (tk, tk_prods') prods;
in (prods', true) end;
val (nt_prods', changed) = (nt_prods, false)
|> nt = lhs ? tokens_fold store start_tks';
val _ =
if not changed andalso tokens_is_empty new_tks then ()
else
Array.update
(prods, nt, ((old_nts, tokens_merge (old_tks, new_tks)), nt_prods'));
in add_tks nts (tokens_add (nt, new_tks) added) end;
val _ = nts_fold add_nts start_nts true;
in add_tks (chains_all_succs chains' [lhs]) [] end;
(*associate productions with new lookaheads*)
val _ =
let
(*propagate added start tokens*)
fun add_starts [] = ()
| add_starts ((changed_nt, new_tks) :: starts) =
let
(*token under which old productions which
depend on changed_nt could be stored*)
val key =
tokens_find (not o tokens_member new_tks) (starts_for_nt changed_nt)
|> the_default unknown_start;
(*copy productions whose lookahead depends on changed_nt;
if key = SOME unknown_start then tk_prods is used to hold
the productions not copied*)
fun update_prods [] result = result
| update_prods ((p as (rhs, _: string, _: nt)) :: ps)
(tk_prods, nt_prods) =
let
(*lookahead dependency for production*)
val (tk, depends) = lookahead_dependency lambdas' rhs nts_empty;
(*test if this production has to be copied*)
val update = nts_member depends changed_nt;
(*test if production could already be associated with
a member of new_tks*)
val lambda =
not (nts_is_unique depends) orelse
not (nts_is_empty depends) andalso is_some tk
andalso tokens_member new_tks (the tk);
(*associate production with new starting tokens*)
fun copy tk nt_prods =
let
val tk_prods = prods_lookup nt_prods tk;
val tk_prods' =
if not lambda then p :: tk_prods
else insert (op =) p tk_prods;
(*if production depends on lambda NT we
have to look for duplicates*)
in prods_update (tk, tk_prods') nt_prods end;
val result =
if update then (tk_prods, tokens_fold copy new_tks nt_prods)
else if key = unknown_start then (p :: tk_prods, nt_prods)
else (tk_prods, nt_prods);
in update_prods ps result end;
(*copy existing productions for new starting tokens*)
fun process_nts nt =
let
val ((nts, tks), nt_prods) = Array.sub (prods, nt);
val tk_prods = prods_lookup nt_prods key;
(*associate productions with new lookahead tokens*)
val (tk_prods', nt_prods') = update_prods tk_prods ([], nt_prods);
val nt_prods'' =
if key = unknown_start then
prods_update (key, tk_prods') nt_prods'
else nt_prods';
val added_tks = tokens_subtract tks new_tks;
val tks' = tokens_merge (tks, added_tks);
val _ = Array.update (prods, nt, ((nts, tks'), nt_prods''));
in tokens_add (nt, added_tks) end;
val ((dependent, _), _) = Array.sub (prods, changed_nt);
in add_starts (starts @ nts_fold process_nts dependent []) end;
in add_starts added_starts' end;
in (chains', lambdas') end;
(* pretty_gram *)
fun pretty_gram (Gram {tags, prods, chains, ...}) =
let
val print_nt = tags_name tags;
fun print_pri p = if p < 0 then "" else Symbol.make_sup ("(" ^ string_of_int p ^ ")");
fun pretty_symb (Terminal (Lexicon.Token (kind, s, _))) =
if kind = Lexicon.Literal then Pretty.quote (Pretty.keyword1 s) else Pretty.str s
| pretty_symb (Nonterminal (tag, p)) = Pretty.str (print_nt tag ^ print_pri p);
fun pretty_const "" = []
| pretty_const c = [Pretty.str ("\<^bold>\<Rightarrow> " ^ quote c)];
fun prod_of_chain from = ([Nonterminal (from, ~1)], "", ~1);
fun pretty_prod (name, tag) =
(prods_content (#2 (Vector.sub (prods, tag))) @ map prod_of_chain (chains_preds chains tag))
|> map (fn (symbs, const, p) =>
Pretty.block (Pretty.breaks
(Pretty.str (name ^ print_pri p ^ " =") :: map pretty_symb symbs @ pretty_const const)));
in maps pretty_prod (tags_content tags) end;
(** operations on grammars **)
val empty_gram =
Gram
{nt_count = 0,
prod_count = 0,
tags = tags_empty,
chains = chains_empty,
lambdas = nts_empty,
prods = Vector.fromList [nt_gram_empty]};
(*Add productions to a grammar*)
fun extend_gram [] gram = gram
| extend_gram xprods (Gram {nt_count, prod_count, tags, chains, lambdas, prods}) =
let
(*Get tag for existing nonterminal or create a new one*)
fun get_tag nt_count tags nt =
(case tags_lookup tags nt of
SOME tag => (nt_count, tags, tag)
| NONE => (nt_count + 1, tags_insert (nt, nt_count) tags, nt_count));
(*Convert symbols to the form used by the parser;
delimiters and predefined terms are stored as terminals,
nonterminals are converted to integer tags*)
fun symb_of [] nt_count tags result = (nt_count, tags, rev result)
| symb_of (Syntax_Ext.Delim s :: ss) nt_count tags result =
symb_of ss nt_count tags
(Terminal (Lexicon.Token (Lexicon.Literal, s, Position.no_range)) :: result)
| symb_of (Syntax_Ext.Argument (s, p) :: ss) nt_count tags result =
let
val (nt_count', tags', new_symb) =
(case Lexicon.predef_term s of
NONE =>
let val (nt_count', tags', s_tag) = get_tag nt_count tags s;
in (nt_count', tags', Nonterminal (s_tag, p)) end
| SOME tk => (nt_count, tags, Terminal tk));
in symb_of ss nt_count' tags' (new_symb :: result) end
| symb_of (_ :: ss) nt_count tags result = symb_of ss nt_count tags result;
(*Convert list of productions by invoking symb_of for each of them*)
fun prod_of [] nt_count prod_count tags result =
(nt_count, prod_count, tags, result)
| prod_of (Syntax_Ext.XProd (lhs, xsymbs, const, pri) :: ps)
nt_count prod_count tags result =
let
val (nt_count', tags', lhs_tag) = get_tag nt_count tags lhs;
val (nt_count'', tags'', prods) = symb_of xsymbs nt_count' tags' [];
in
prod_of ps nt_count'' (prod_count + 1) tags''
((lhs_tag, (prods, const, pri)) :: result)
end;
val (nt_count', prod_count', tags', xprods') =
prod_of xprods nt_count prod_count tags [];
(*Copy array containing productions of old grammar;
this has to be done to preserve the old grammar while being able
to change the array's content*)
val prods' =
let
fun get_prod i =
if i < nt_count then Vector.sub (prods, i)
else nt_gram_empty;
in Array.tabulate (nt_count', get_prod) end;
val (chains', lambdas') =
(chains, lambdas) |> fold (add_production prods') xprods';
in
Gram
{nt_count = nt_count',
prod_count = prod_count',
tags = tags',
chains = chains',
lambdas = lambdas',
prods = Array.vector prods'}
end;
fun make_gram xprods = extend_gram xprods empty_gram;
(** parser **)
(* parsetree *)
datatype parsetree =
Node of string * parsetree list |
Tip of Lexicon.token;
exception PARSETREE of parsetree;
fun pretty_parsetree parsetree =
let
fun pretty (Node (c, pts)) =
[Pretty.enclose "(" ")" (Pretty.breaks (Pretty.quote (Pretty.str c) :: maps pretty pts))]
| pretty (Tip tok) =
if Lexicon.valued_token tok then [Pretty.str (Lexicon.str_of_token tok)] else [];
in (case pretty parsetree of [prt] => prt | _ => raise PARSETREE parsetree) end;
(* parser state *)
type state =
nt * int * (*identification and production precedence*)
parsetree list * (*already parsed nonterminals on rhs*)
symb list * (*rest of rhs*)
string * (*name of production*)
int; (*index for previous state list*)
(*Get all rhss with precedence >= min_prec*)
fun get_RHS min_prec = filter (fn (_, _, prec: int) => prec >= min_prec);
(*Get all rhss with precedence >= min_prec and < max_prec*)
fun get_RHS' min_prec max_prec =
filter (fn (_, _, prec: int) => prec >= min_prec andalso prec < max_prec);
(*Make states using a list of rhss*)
fun mk_states i lhs_ID rhss =
let fun mk_state (rhs, id, prod_prec) = (lhs_ID, prod_prec, [], rhs, id, i);
in map mk_state rhss end;
(*Add parse tree to list and eliminate duplicates
saving the maximum precedence*)
fun conc (t: parsetree list, prec: int) [] = (NONE, [(t, prec)])
| conc (t, prec) ((t', prec') :: ts) =
if t = t' then
(SOME prec',
if prec' >= prec then (t', prec') :: ts
else (t, prec) :: ts)
else
let val (n, ts') = conc (t, prec) ts
in (n, (t', prec') :: ts') end;
(*Update entry in used*)
fun update_trees (A, t) used =
let
val (i, ts) = the (Inttab.lookup used A);
val (n, ts') = conc t ts;
in (n, Inttab.update (A, (i, ts')) used) end;
(*Replace entry in used*)
fun update_prec (A, prec) =
Inttab.map_entry A (fn (_, ts) => (prec, ts));
fun getS A max_prec NONE Si =
filter
(fn (_, _, _, Nonterminal (B, prec) :: _, _, _) => A = B andalso prec <= max_prec
| _ => false) Si
| getS A max_prec (SOME min_prec) Si =
filter
(fn (_, _, _, Nonterminal (B, prec) :: _, _, _) =>
A = B andalso prec > min_prec andalso prec <= max_prec
| _ => false) Si;
fun get_states Estate j A max_prec =
filter
(fn (_, _, _, Nonterminal (B, prec) :: _, _, _) => A = B andalso prec <= max_prec
| _ => false)
(Array.sub (Estate, j));
fun movedot_term c (A, j, ts, Terminal a :: sa, id, i) =
if Lexicon.valued_token c orelse id <> ""
then (A, j, Tip c :: ts, sa, id, i)
else (A, j, ts, sa, id, i);
fun movedot_nonterm tt (A, j, ts, Nonterminal _ :: sa, id, i) =
(A, j, tt @ ts, sa, id, i);
fun movedot_lambda [] _ = []
| movedot_lambda ((t, ki) :: ts) (state as (B, j, tss, Nonterminal (A, k) :: sa, id, i)) =
if k <= ki then (B, j, t @ tss, sa, id, i) :: movedot_lambda ts state
else movedot_lambda ts state;
(*trigger value for warnings*)
val branching_level =
Config.int (Config.declare ("syntax_branching_level", \<^here>) (fn _ => Config.Int 600));
(*get all productions of a NT and NTs chained to it which can
be started by specified token*)
fun prods_for (Gram {prods, chains, ...}) tk nt =
let
fun token_prods prods =
fold cons (prods_lookup prods tk) #>
fold cons (prods_lookup prods Lexicon.dummy);
fun nt_prods nt = #2 (Vector.sub (prods, nt));
in fold (token_prods o nt_prods) (chains_all_preds chains [nt]) [] end;
fun PROCESSS gram Estate i c states =
let
fun processS _ [] (Si, Sii) = (Si, Sii)
| processS used (S :: States) (Si, Sii) =
(case S of
(_, _, _, Nonterminal (nt, min_prec) :: _, _, _) =>
let (*predictor operation*)
val (used', new_states) =
(case Inttab.lookup used nt of
SOME (used_prec, l) => (*nonterminal has been processed*)
if used_prec <= min_prec then
(*wanted precedence has been processed*)
(used, movedot_lambda l S)
else (*wanted precedence hasn't been parsed yet*)
let
val tk_prods = prods_for gram c nt;
val States' =
mk_states i nt (get_RHS' min_prec used_prec tk_prods);
in (update_prec (nt, min_prec) used, movedot_lambda l S @ States') end
| NONE => (*nonterminal is parsed for the first time*)
let
val tk_prods = prods_for gram c nt;
val States' = mk_states i nt (get_RHS min_prec tk_prods);
in (Inttab.update (nt, (min_prec, [])) used, States') end);
in
processS used' (new_states @ States) (S :: Si, Sii)
end
| (_, _, _, Terminal a :: _, _, _) => (*scanner operation*)
processS used States
(S :: Si,
if Lexicon.tokens_match_ord (a, c) = EQUAL then movedot_term c S :: Sii else Sii)
| (A, prec, ts, [], id, j) => (*completer operation*)
let val tt = if id = "" then ts else [Node (id, rev ts)] in
if j = i then (*lambda production?*)
let
val (prec', used') = update_trees (A, (tt, prec)) used;
val Slist = getS A prec prec' Si;
val States' = map (movedot_nonterm tt) Slist;
in processS used' (States' @ States) (S :: Si, Sii) end
else
let val Slist = get_states Estate j A prec
in processS used (map (movedot_nonterm tt) Slist @ States) (S :: Si, Sii) end
end)
in processS Inttab.empty states ([], []) end;
fun produce gram stateset i indata prev_token =
(case Array.sub (stateset, i) of
[] =>
let
val toks = if Lexicon.is_eof prev_token then indata else prev_token :: indata;
val pos = Position.here (Lexicon.pos_of_token prev_token);
in
if null toks then
error ("Inner syntax error: unexpected end of input" ^ pos)
else
error ("Inner syntax error" ^ pos ^
Markup.markup Markup.no_report
("\n" ^ Pretty.string_of
(Pretty.block [
Pretty.str "at", Pretty.brk 1,
Pretty.block
(Pretty.str "\"" ::
Pretty.breaks (map (Pretty.str o Lexicon.str_of_token) (#1 (split_last toks))) @
[Pretty.str "\""])])))
end
| s =>
(case indata of
[] => s
| c :: cs =>
let
val (si, sii) = PROCESSS gram stateset i c s;
val _ = Array.update (stateset, i, si);
val _ = Array.update (stateset, i + 1, sii);
in produce gram stateset (i + 1) cs c end));
fun get_trees states = map_filter (fn (_, _, [pt], _, _, _) => SOME pt | _ => NONE) states;
fun earley (gram as Gram {tags, ...}) startsymbol indata =
let
val start_tag =
(case tags_lookup tags startsymbol of
SOME tag => tag
| NONE => error ("Inner syntax: bad grammar root symbol " ^ quote startsymbol));
val S0 = [(~1, 0, [], [Nonterminal (start_tag, 0), Terminal Lexicon.eof], "", 0)];
val s = length indata + 1;
val Estate = Array.array (s, []);
val _ = Array.update (Estate, 0, S0);
in
get_trees (produce gram Estate 0 indata Lexicon.eof)
end;
fun parse gram start toks =
let
val end_pos =
(case try List.last toks of
NONE => Position.none
| SOME tok => Lexicon.end_pos_of_token tok);
val r =
(case earley gram start (toks @ [Lexicon.mk_eof end_pos]) of
[] => raise Fail "Inner syntax: no parse trees"
| pts => pts);
in r end;
fun guess_infix_lr (Gram gram) c = (*based on educated guess*)
let
fun freeze a = map_range (curry Vector.sub a) (Vector.length a);
val prods = maps (prods_content o #2) (freeze (#prods gram));
fun guess (SOME ([Nonterminal (_, k),
Terminal (Lexicon.Token (Lexicon.Literal, s, _)), Nonterminal (_, l)], _, j)) =
if k = j andalso l = j + 1 then SOME (s, true, false, j)
else if k = j + 1 then if l = j then SOME (s, false, true, j)
else if l = j + 1 then SOME (s, false, false, j)
else NONE
else NONE
| guess _ = NONE;
in guess (find_first (fn (_, s, _) => s = c) prods) end;
end;