Hi Vincent, > Le 30 juin 2020 à 06:47, Akim Demaille <[email protected]> a écrit : > > Hi Vincent, > >> Le 28 juin 2020 à 21:30, Vincent Imbimbo <[email protected]> a écrit : >> >> Hey Akim, >> I think I found a good place for the cex documentation in bison.texi, but it >> might still be a bit awkward. Tell me if you have any ideas for changes. > > Thanks a lot for this! I made a few changes (e.g., the name is > -Wcounterexamples, plural, moving -Wcex as foremost feature of 3.7, wrap at > 76 columns, remove the duplicate example in NEWS which is B&W, sort > references, etc.) > > Here's how I installed your commit. I'll proceed from there. Thanks!
I believe your work deserve a lot more space than what you did. Here is my shot. I would be very happy to get feedback, especially from you Vincent. Cheers! commit cf1f69b98723fab92c6b2c7be5c678e7cd0a5a80 Author: Akim Demaille <[email protected]> Date: Thu Jul 2 20:35:33 2020 +0200 cex: give more details about -Wcex and -rcex * data/bison-default.css: Cobalt does not seem to be supported. * doc/bison.texi (Counterexamples): A new section. (Understanding): Show the counterexamples as it shows in the report: with its items. (Bison Options): Document -Wcex and -rcex. diff --git a/data/bison-default.css b/data/bison-default.css index 2652e03f..5248c294 100644 --- a/data/bison-default.css +++ b/data/bison-default.css @@ -44,7 +44,7 @@ .cex-0 { color: yellow; } .cex-1 { color: green; } .cex-2 { color: blue; } -.cex-3 { color: cobalt; } +.cex-3 { color: purple; } .cex-4 { color: violet; } .cex-5 { color: orange; } .cex-6 { color: brown; } diff --git a/doc/bison.texi b/doc/bison.texi index b45bbca9..c4a60ba9 100644 --- a/doc/bison.texi +++ b/doc/bison.texi @@ -53,11 +53,11 @@ \gdef\colorBlue{% \setcolor{\rgbBlue}% } - -\gdef\rgbWarning{0.50 0 0.50} -\gdef\colorWarning{% - \setcolor{\rgbWarning}% +\gdef\rgbPurple{0.50 0 0.50} +\gdef\colorPurple{% + \setcolor{\rgbPurple}% } + \gdef\rgbError{0.80 0 0} \gdef\colorError{% \setcolor{\rgbError}% @@ -84,10 +84,10 @@ @macro colorBlue @inlineraw{html, <b style="color:blue">} @end macro - -@macro colorWarning +@macro colorPurple @inlineraw{html, <b style="color:darkviolet">} @end macro + @macro colorError @inlineraw{html, <b style="color:red">} @end macro @@ -115,8 +115,12 @@ @colorBlue{}\text\@colorOff{} @end macro +@macro purple{text} +@colorPurple{}\text\@colorOff{} +@end macro + @macro dwarning{text} -@colorWarning{}\text\@colorOff{} +@purple{\text\} @end macro @macro derror{text} @@ -9824,7 +9828,158 @@ and understand the parser run-time traces (@pxref{Tracing}). @node Counterexamples @section Generation of Counterexamples -@fixme +Solving conflicts is probably the most delicated part of the design of an LR +parser, as demonstrated by the number of sections devoted to them in this +very documentation. To solve a conflict, one must understand it: when does +it occur? Is it because of a flaw in the grammar? Is it rather because +LR(1) cannot cope with this grammar? + +On difficulty is that conflicts occur in the @emph{automaton}, and it can be +tricky to related them to issues in the @emph{grammar} itself. With +experience and patience, analysis the detailed description of the automaton +(@pxref{Understanding}) allows to find example strings that reach these conflicts. + +That task is made much easier thanks to the generation of counterexamples, +initially developed by Chinawat Isradisaikul and Andrew Myers +@pcite{Isradisaikul 2015}. + +As a first example, see the example grammar of @ref{Shift/Reduce}, which +features on shift/reduce conflict: + +@example +$ @kbd{bison if-then-else.y} +if-then-else.y: warning: 1 shift/reduce conflict [-Wconflicts-sr] +if-then-else.y: warning: rerun with option '-Wcounterexamples' to generate conflict counterexamples [-Wother] +@end example + +@noindent +Let's rerun @command{bison} with the option +@option{-Wcex}/@option{-Wcounterexamples}@inlinefmt{info, (the following +output is actually in color)}: + +@ifhtml +@example +Shift/reduce conflict on token "else": +@group + Example @yellow{"if" expr "then"} @blue{"if" expr "then" stmt} @red{•} @yellow{"else" stmt} + First derivation @yellow{if_stmt ::=[ "if" expr "then"} @green{stmt ::=[} @blue{if_stmt ::=[ "if" expr "then" stmt} @red{•} @blue{]} @green{]} @yellow{"else" stmt ]} + Example @yellow{"if" expr "then"} @blue{"if" expr "then" stmt} @red{•} @blue{"else" stmt} + Second derivation @yellow{if_stmt ::=[ "if" expr "then"} @green{stmt ::=[} @blue{if_stmt ::=[ "if" expr "then" stmt} @red{•} @blue{"else" stmt ]} @green{]} @yellow{]} +@end group +@end example +@end ifhtml +@ifnothtml +@smallexample +Shift/reduce conflict on token "else": +@group + Example + @yellow{"if" expr "then"} @blue{"if" expr "then" stmt} @red{•} @yellow{"else" stmt} + First derivation + @yellow{if_stmt ::=[ "if" expr "then"} @green{stmt ::=[} @blue{if_stmt ::=[ "if" expr "then" stmt} @red{•} @blue{]} @green{]} @yellow{"else" stmt ]} + Example + @yellow{"if" expr "then"} @blue{"if" expr "then" stmt} @red{•} @blue{"else" stmt} + Second derivation + @yellow{if_stmt ::=[ "if" expr "then"} @green{stmt ::=[} @blue{if_stmt ::=[ "if" expr "then" stmt} @red{•} @blue{"else" stmt ]} @green{]} @yellow{]} +@end group +@end smallexample +@end ifnothtml + +This shows two different derivations for one single expression. That +demonstrates that the grammar is ambiguous. + +@sp 1 + +As a more delicate example, consider the example grammar of +@ref{Reduce/Reduce}, which features a reduce/reduce conflict: + +@example +%% +sequence: + %empty +| maybeword +| sequence "word" +; +maybeword: + %empty +| "word" +; +@end example + +Bison generates the following counterexamples: + +@example +$ @kbd{bison -Wcex sequence.y} +sequence.y: @dwarning{warning}: 1 shift/reduce conflict [@dwarning{-Wconflicts-sr}] +sequence.y: @dwarning{warning}: 2 reduce/reduce conflicts [@dwarning{-Wconflicts-rr}] +Shift/reduce conflict on token "word": + Example @red{•} @yellow{"word"} + First derivation @yellow{sequence ::=[} @green{sequence ::=[} @red{•} @green{]} @yellow{"word" ]} + Example @red{•} @green{"word"} + Second derivation @yellow{sequence ::=[} @green{maybeword ::=[} @red{•} @green{"word" ]} @yellow{]} + +Reduce/reduce conflict on tokens $end, "word": + Example @red{•} + First derivation @yellow{sequence ::=[} @red{•} @yellow{]} + Example @red{•} + Second derivation @yellow{sequence ::=[} @green{maybeword ::=[} @red{•} @green{]} @yellow{]} + +Shift/reduce conflict on token "word": + Example @red{•} @yellow{"word"} + First derivation @yellow{sequence ::=[} @green{sequence ::=[} @blue{maybeword ::=[} @red{•} @blue{]} @green{]} @yellow{"word" ]} + Example @red{•} @green{"word"} + Second derivation @yellow{sequence ::=[} @green{maybeword ::=[} @red{•} @green{"word" ]} @yellow{]} + +sequence.y:8.3-45: @dwarning{warning}: rule useless in parser due to conflicts [@dwarning{-Wother}] + 8 | @dwarning{%empty @{ printf ("empty maybeword\n"); @}} + | @dwarning{^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~} +@end example + +Each of these three conflicts, again, prove that the grammar is ambiguous. +For instance, the second conflict (the reduce/reduce one) shows that the +grammar accept the empty input in two different ways. + +@sp 1 + +Sometimes, the search will not find an example that can be derived in two +ways. In these cases, counterexample generation will provide two examples +that are the same up until the dot. Most notably, this will happen when +your grammar requires a stronger parser (more lookahead, LR instead of +LALR). The following example isn't LR(1): + +@example +%token ID +%% +s: a ID +a: expr +expr: %empty | expr ID ',' +@end example + +@command{bison} reports: + +@smallexample +Shift/reduce conflict on token ID: + First example @blue{expr} @red{•} @green{ID} @yellow{$end} + First derivation @yellow{$accept ::=[} @green{s ::=[} @blue{a ::=[ expr} @red{•} @blue{]} @green{ID ]} @yellow{$end ]} + Second example @purple{expr} @red{•} @purple{ID ','} @green{ID} @yellow{$end} + Second derivation @yellow{$accept ::=[} @green{s ::=[} @blue{a ::=[} @purple{expr ::=[ expr} @red{•} @purple{ID ',' ]} @blue{]} @green{ID ]} @yellow{$end ]} +@end smallexample + +This conflict is caused by the parser not having enough information to know +the difference between these two examples. The parser would need an +additional lookahead token to know whether or not a comma follows the +@code{ID} after @code{expr}. These types of conflicts tend to be more +difficult to fix, and usually need a rework of the grammar. In this case, +it can be fixed by changing around the recursion: @code{expr: ID | ',' expr +ID}. + +Alternatively, you might also want to consider using a GLR parser +(@pxref{GLR Parsers}). + +@sp 1 + +On occasions, it is useful to look at counterexamples @emph{in situ}: with +the automaton report (@xref{Understanding}, in particular @ref{state-8,, +State 8}). @node Understanding @section Understanding Your Parser @@ -9889,62 +10044,6 @@ calc.y: @dwarning{warning}: 7 shift/reduce conflicts [@dwarning{-Wconflicts-sr}] calc.y: @dwarning{warning}: rerun with option '-Wcounterexamples' to generate conflict counterexamples [@dwarning{-Wother}] @end smallexample -When given @option{-Wcounterexamples}, @command{bison} will run a search for -strings in your grammar that better demonstrate you -conflicts. Counterexample generation was initially developed by Chinawat -Isradisaikul and Andrew Myers @pcite{Isradisaikul 2015}. For -@file{calc.y}, the first printed example is: - -@example -Shift/reduce conflict on token '/': - Example @green{exp '+' exp} @red{•} @yellow{'/' exp} - First derivation @yellow{exp ::=[} @green{exp ::=[ exp '+' exp} @red{•} @green{]} @yellow{'/' exp ]} - Example @yellow{exp '+'} @green{exp} @red{•} @green{'/' exp} - Second derivation @yellow{exp ::=[ exp '+'} @green{exp ::=[ exp} @red{•} @green{'/' exp ]} @yellow{]} -@end example - -This shows two separate derivations in the grammar for the same @code{exp}: -@samp{e1 + e2 / e3}. The derivations show how your rules would parse the -given example. Here, the first derivation completes a reduction when seeing -@samp{/}, causing @samp{e1 + e2} to be grouped as an @code{exp}. The second -derivation shifts on @samp{/}, resulting in @samp{e2 / e3} being grouped as -an @code{exp}. Therefore, it is easy to see that adding @code{%precedence} -directives would fix this conflict. - -Sometimes, the search will not find an example that can be derived in two -ways. In these cases, counterexample generation will provide two examples -that are the same up until the dot. Most notably, this will happen when -your grammar requires a stronger parser (more lookahead, LR instead of -LALR). The following example isn't LR(1): - -@example -%token ID -%% -s: a ID -a: expr -expr: %empty | expr ID ',' -@end example - -@command{bison} reports: - -@smallexample -Shift/reduce conflict on token ID: - First example @blue{expr} @red{•} @green{ID} @yellow{$end} - First derivation @yellow{$accept ::=[} @green{s ::=[} @blue{a ::=[ expr} @red{•} @blue{]} @green{ID ]} @yellow{$end ]} - Second example @blue{expr} @red{•} @blue{ID ','} @green{ID} @yellow{$end} - Second derivation @yellow{$accept ::=[} @green{s ::=[} @blue{a ::=[ expr ::=[ expr} @red{•} @blue{ID ',' ] ]} @green{ID ]} @yellow{$end ]} -@end smallexample - -This conflict is caused by the parser not having enough information to know -the difference between these two examples. The parser would need an -additional lookahead token to know whether or not a comma follows the -@code{ID} after @code{expr}. These types of conflicts tend to be more -difficult to fix, and usually need a rework of the grammar. In this case, -it can be fixed by changing around the recursion: @code{expr: ID | ',' expr -ID}. - -Counterexamples can also be written to a file with @option{--report=cex}. - Going back to the calc example, when given @option{--report=state}, in addition to @file{calc.tab.c}, it creates a file @file{calc.output} with contents detailed below. The order of the output and the exact @@ -10176,6 +10275,7 @@ State 7 exp go to state 11 @end example +@anchor{state-8} As was announced in beginning of the report, @samp{State 8 conflicts: 1 shift/reduce}: @@ -10249,6 +10349,27 @@ Conflict between rule 1 and token '-' resolved as reduce (%left '-'). Conflict between rule 1 and token '*' resolved as shift ('+' < '*'). @end example +When given @option{--report=counterexamples}, @command{bison} will generate +counterexamples within the report, augmented with the corresponding items +(@pxref{Counterexamples}). + +@example +Shift/reduce conflict on token '/': + 1 exp: exp '+' exp • + 4 exp: exp • '/' exp + Example @green{exp '+' exp} @red{•} @yellow{'/' exp} + First derivation @yellow{exp ::=[} @green{exp ::=[ exp '+' exp} @red{•} @green{]} @yellow{'/' exp ]} + Example @yellow{exp '+'} @green{exp} @red{•} @green{'/' exp} + Second derivation @yellow{exp ::=[ exp '+'} @green{exp ::=[ exp} @red{•} @green{'/' exp ]} @yellow{]} +@end example + +This shows two separate derivations in the grammar for the same @code{exp}: +@samp{e1 + e2 / e3}. The derivations show how your rules would parse the +given example. Here, the first derivation completes a reduction when seeing +@samp{/}, causing @samp{e1 + e2} to be grouped as an @code{exp}. The second +derivation shifts on @samp{/}, resulting in @samp{e2 / e3} being grouped as +an @code{exp}. Therefore, it is easy to see that adding @code{%precedence} +directives would fix this conflict. The remaining states are similar: @@ -11096,6 +11217,13 @@ unexpected number of conflicts is an error, and an expected number of conflicts is not reported, so @option{-W} and @option{--warning} then have no effect on the conflict report. +@item counterexamples +@itemx cex +Provide counterexamples for conflicts. @xref{Counterexamples}. +Counterexamples take time to compute. The option @option{-Wcex} should be +used by the developer when working on the grammar; it hardly makes sense to +use it in a CI. + @item dangling-alias Report string literals that are not bound to a token symbol. @@ -11470,6 +11598,13 @@ each rule's lookahead set. Implies @code{state}. Explain how conflicts were solved thanks to precedence and associativity directives. +@item counterexamples +@itemx cex +Look for counterexamples for the conflicts. @xref{Counterexamples}. +Counterexamples take time to compute. The option @option{-rcex} should be +used by the developer when working on the grammar; it hardly makes sense to +use it in a CI. + @item all Enable all the items.
