X-Git-Url: https://git.saurik.com/bison.git/blobdiff_plain/297e263a0050959e0fd139ad66e71383fc9ac4db..3fd1d6b2f36722c64f8db77400c6f6a2639ebae0:/cfg.mk diff --git a/cfg.mk b/cfg.mk index 392d2112..ea9806fd 100644 --- a/cfg.mk +++ b/cfg.mk @@ -24,6 +24,7 @@ regen: _version # Used in maint.mk's web-manual rule manual_title = The Yacc-compatible Parser Generator +gendocs_options_ = -I $(abs_top_srcdir)/doc -I $(abs_top_builddir)/doc # It's useful to run maintainer-*check* targets during development, but we # don't want to wait on a recompile because of an update to $(VERSION). Thus, @@ -54,17 +55,126 @@ update-copyright: update-b4-copyright update-package-copyright-year update-copyright-env = \ UPDATE_COPYRIGHT_FORCE=1 UPDATE_COPYRIGHT_USE_INTERVALS=1 -exclude = $(foreach a,$(1),$(eval exclude_file_name_regexp--sc_$(a))) + +## -------------------- ## +## More syntax-checks. ## +## -------------------- ## + +# At least for Mac OS X's grep, the order between . and [ in "[^.[]" +# matters: +# $ LC_ALL=fr_FR grep -nE '[^[.]' /dev/null +# $ LC_ALL=C grep -nE '[^[.]' /dev/null +# grep: invalid collating element or class +# $ LC_ALL=fr_FR grep -nE '[^.[]' /dev/null +# $ LC_ALL=C grep -nE '[^.[]' /dev/null +sc_at_parser_check: + @prohibit='AT_PARSER_CHECK\(\[+[^.[]|AT_CHECK\(\[+\./' \ + halt='use AT_PARSER_CHECK for and only for generated parsers' \ + $(_sc_search_regexp) + +# Indent only with spaces. +# Taken from Coreutils. +sc_prohibit_tab_based_indentation: + @prohibit='^ * ' \ + halt='TAB in indentation; use only spaces' \ + $(_sc_search_regexp) + +# Prohibit the use of `...` in tests/. Use $(...) instead. +# Taken from Coreutils. +# Not ready for Bison yet. +#sc_prohibit_test_backticks: +# @prohibit='`' in_vc_files='^tests/' \ +# halt='use $$(...), not `...` in tests/' \ +# $(_sc_search_regexp) + +# Enforce recommended preprocessor indentation style. +# Taken from Coreutils. +sc_preprocessor_indentation: + @if cppi --version >/dev/null 2>&1; then \ + $(VC_LIST_EXCEPT) | grep '\.[ch]$$' | xargs cppi -a -c \ + || { echo '$(ME): incorrect preprocessor indentation' 1>&2; \ + exit 1; }; \ + else \ + echo '$(ME): skipping test $@: cppi not installed' 1>&2; \ + fi + +########################################################### +# Taken from Coreutils. +_p0 = \([^"'/]\|"\([^\"]\|[\].\)*"\|'\([^\']\|[\].\)*' +_pre = $(_p0)\|[/][^"'/*]\|[/]"\([^\"]\|[\].\)*"\|[/]'\([^\']\|[\].\)*'\)* +_pre_anchored = ^\($(_pre)\) +_comment_and_close = [^*]\|[*][^/*]\)*[*][*]*/ +# help font-lock mode: ' + +# A sed expression that removes ANSI C and ISO C99 comments. +# Derived from the one in GNU gettext's 'moopp' preprocessor. +_sed_remove_comments = \ +/[/][/*]/{ \ + ta; \ + :a; \ + s,$(_pre_anchored)//.*,\1,; \ + te; \ + s,$(_pre_anchored)/[*]\($(_comment_and_close),\1 ,; \ + ta; \ + /^$(_pre)[/][*]/{ \ + s,$(_pre_anchored)/[*].*,\1 ,; \ + tu; \ + :u; \ + n; \ + s,^\($(_comment_and_close),,; \ + tv; \ + s,^.*$$,,; \ + bu; \ + :v; \ + }; \ + :e; \ +} +# Quote all single quotes. +_sed_rm_comments_q = $(subst ','\'',$(_sed_remove_comments)) +# help font-lock mode: ' + +_space_before_paren_exempt =? \\n\\$$ +_space_before_paren_exempt = \ + (^ *\#|\\n\\$$|%s\(to %s|(date|group|character)\(s\)) +# Ensure that there is a space before each open parenthesis in C code. +sc_space_before_open_paren: + @if $(VC_LIST_EXCEPT) | grep -l '\.[ch]$$' > /dev/null; then \ + fail=0; \ + for c in $$($(VC_LIST_EXCEPT) | grep '\.[ch]$$'); do \ + sed '$(_sed_rm_comments_q)' $$c 2>/dev/null \ + | grep -i '[[:alnum:]](' \ + | grep -vE '$(_space_before_paren_exempt)' \ + | grep . && { fail=1; echo "*** $$c"; }; \ + done; \ + test $$fail = 1 && \ + { echo '$(ME): the above files lack a space-before-open-paren' \ + 1>&2; exit 1; } || :; \ + else :; \ + fi + +## -------------------------- ## +## syntax-checks exceptions. ## +## -------------------------- ## + +exclude = \ + $(foreach a,$(1),$(eval $(subst $$,$$$$,exclude_file_name_regexp--sc_$(a)))) $(call exclude, \ bindtextdomain=^lib/main.c$$ \ + preprocessor_indentation=^data/|^lib/|^src/parse-gram.[ch]$$ \ program_name=^lib/main.c$$ \ - prohibit_always-defined_macros=^data/yacc.c|^djgpp/ \ + prohibit_always-defined_macros=^data/yacc.c$$|^djgpp/ \ prohibit_always-defined_macros+=?|^lib/timevar.c$$ \ prohibit_always-defined_macros+=?|^src/(parse-gram.c|system.h)$$ \ prohibit_always-defined_macros+=?|^tests/regression.at$$ \ - prohibit_empty_lines_at_EOF=^src/parse-gram.h$$ \ - prohibit_strcmp=^doc/bison\.texinfo$$ \ + prohibit_always_true_header_tests=^djgpp/subpipe.h$$|^lib/timevar.c$$ \ + prohibit_always_true_header_tests+=?|^m4/timevar.m4$$ \ + prohibit_defined_have_decl_tests=?|^lib/timevar.c$$ \ + prohibit_doubled_word=^tests/named-refs.at$$ \ + prohibit_magic_number_exit=^doc/bison.texi$$ \ + prohibit_magic_number_exit+=?|^tests/(conflicts|regression).at$$ \ + prohibit_strcmp=^doc/bison\.texi$$ \ + prohibit_tab_based_indentation=\.(am|mk)$$|^djgpp/|^\.git \ require_config_h_first=^(lib/yyerror|data/(glr|yacc))\.c$$ \ space_tab=^tests/(input|c\+\+)\.at$$ \ - unmarked_diagnostics=^djgpp/ \ + unmarked_diagnostics=^(djgpp/|doc/bison.texi$$|tests/c\+\+\.at$$) \ )