* Re: coreutils and GCC -fanalyzer
[not found] ` <3a074086-47f7-638f-50a8-cd7ceb46a11a@draigBrady.com>
@ 2020-05-23 16:08 ` Paul Eggert
2020-05-23 16:22 ` Tim Rühsen
2020-07-02 0:26 ` Paul Eggert
0 siblings, 2 replies; 7+ messages in thread
From: Paul Eggert @ 2020-05-23 16:08 UTC (permalink / raw)
To: Pádraig Brady; +Cc: Gnulib bugs, coreutils
On 5/23/20 8:56 AM, Pádraig Brady wrote:
> Note -Wanalyzer... flags are ineffective without -fanalyzer.
> Also -Wanalyzer-too-complex (now auto enabled by gnulib) is currently quite
> problematic with -fanalyzer.
Hmm, that's odd, since I recall getting more warnings with those flags enabled.
Perhaps not all of them are disabled when you don't use -fanalyzer? Or maybe my
memory is faulty.
Anyway, cleary the Gnulib manywarnings module needs to be fixed to do a better
job here, and this reminds me of another problem that has been bothering me for
quite some time: ./configure spends way too much time testing for individual
warnings when the manywarnings module is in use. So I am thinking of killing two
stones by doing the following.
1. Test for -fanalyzer, -Wall, -Wextra.
2. Test for flags not automatically enabled by -fanalyzer, -Wall, -Wextra but
flags that we want anyway.
3. Test for flags automatically enabled by -fanalyzer, -Wall, -Wextra that are
also flags that we don't want.
This should make 'configure' much faster. It'll also fix the issue you
mentioned. And we can make sure that -Wanalyzer-too-complex is not mentioned in
step (2) so that Gnulib won't generate it.
Ccing to bug-gnulib since this is really more of a bug-gnulib issue.
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: coreutils and GCC -fanalyzer
2020-05-23 16:08 ` coreutils and GCC -fanalyzer Paul Eggert
@ 2020-05-23 16:22 ` Tim Rühsen
2020-07-02 0:26 ` Paul Eggert
1 sibling, 0 replies; 7+ messages in thread
From: Tim Rühsen @ 2020-05-23 16:22 UTC (permalink / raw)
To: Paul Eggert, Pádraig Brady; +Cc: Gnulib bugs, coreutils
[-- Attachment #1.1: Type: text/plain, Size: 1675 bytes --]
On 23.05.20 18:08, Paul Eggert wrote:
> On 5/23/20 8:56 AM, Pádraig Brady wrote:
>> Note -Wanalyzer... flags are ineffective without -fanalyzer.
>> Also -Wanalyzer-too-complex (now auto enabled by gnulib) is currently quite
>> problematic with -fanalyzer.
>
> Hmm, that's odd, since I recall getting more warnings with those flags enabled.
> Perhaps not all of them are disabled when you don't use -fanalyzer? Or maybe my
> memory is faulty.
>
> Anyway, cleary the Gnulib manywarnings module needs to be fixed to do a better
> job here, and this reminds me of another problem that has been bothering me for
> quite some time: ./configure spends way too much time testing for individual
> warnings when the manywarnings module is in use. So I am thinking of killing two
> stones by doing the following.
>
> 1. Test for -fanalyzer, -Wall, -Wextra.
>
> 2. Test for flags not automatically enabled by -fanalyzer, -Wall, -Wextra but
> flags that we want anyway.
>
> 3. Test for flags automatically enabled by -fanalyzer, -Wall, -Wextra that are
> also flags that we don't want.
>
> This should make 'configure' much faster. It'll also fix the issue you
> mentioned. And we can make sure that -Wanalyzer-too-complex is not mentioned in
> step (2) so that Gnulib won't generate it.
This has been done for Wget2
https://gitlab.com/gnuwget/wget2/-/blob/master/m4/wget_manywarnings.m4
Example usage at L123+:
https://gitlab.com/gnuwget/wget2/-/blob/master/configure.ac
To make it usable for other projects, it needs some merging with the
manywarnings.m4 from gnulib, IMO. I would support any volunteer as good
as i can.
Regards, Tim
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: coreutils and GCC -fanalyzer
2020-05-23 16:08 ` coreutils and GCC -fanalyzer Paul Eggert
2020-05-23 16:22 ` Tim Rühsen
@ 2020-07-02 0:26 ` Paul Eggert
2020-07-10 21:21 ` Pádraig Brady
1 sibling, 1 reply; 7+ messages in thread
From: Paul Eggert @ 2020-07-02 0:26 UTC (permalink / raw)
To: Pádraig Brady; +Cc: Tim Rühsen, Gnulib bugs, coreutils
[-- Attachment #1: Type: text/plain, Size: 1425 bytes --]
On 5/23/20 9:08 AM, Paul Eggert wrote:
> So I am thinking of killing two
> stones by doing the following.
>
> 1. Test for -fanalyzer, -Wall, -Wextra.
>
> 2. Test for flags not automatically enabled by -fanalyzer, -Wall, -Wextra but
> flags that we want anyway.
>
> 3. Test for flags automatically enabled by -fanalyzer, -Wall, -Wextra that are
> also flags that we don't want.
I did that in Gnulib by installing the attached patch. This could greatly
increase compile times due to the -fanalyzer option, so let's keep an eye out
for that.
On 5/23/20 9:22 AM, Tim Rühsen wrote:
> This has been done for Wget2
> https://gitlab.com/gnuwget/wget2/-/blob/master/m4/wget_manywarnings.m4
>
> Example usage at L123+:
> https://gitlab.com/gnuwget/wget2/-/blob/master/configure.ac
>
> To make it usable for other projects, it needs some merging with the
> manywarnings.m4 from gnulib, IMO. I would support any volunteer as good
> as i can.
Thanks. I think this different approach should be a win. I wanted to get the old
approach working with GCC 10.1 first though (and did that via the attached
patch) before trying out this new method, so that we can verify that switching
from old to new methods improves 'configure' performance without changing the
set of flags given to GCC. Perhaps we can wait for a bit for a shakeout period
for -fanalyzer etc., and then look into switching methods.
[-- Attachment #2: 0001-manywarnings-improve-port-to-GCC-10.1.patch --]
[-- Type: text/x-patch, Size: 24668 bytes --]
From 3831e2e9f355c557b0c0ed9712548b62feaf694f Mon Sep 17 00:00:00 2001
From: Paul Eggert <eggert@cs.ucla.edu>
Date: Wed, 1 Jul 2020 17:00:51 -0700
Subject: [PATCH] manywarnings: improve port to GCC 10.1
* build-aux/gcc-warning.spec: Also list warnings that are default
or are enabled by already-given flags. This lets us speed up
checking for attributes, and makes the generated compilation
commands shorter. Add -Wanalyzer-too-complex (too much noise).
* m4/manywarnings.m4 (gl_MANYWARN_COMPLEMENT)
(gl_MANYWARN_ALL_GCC): Use gl_AS_VAR_APPEND to append
to shell variables that may have long values.
(gl_MANYWARN_ALL_GCC): Omit flags that are default or are
consequences of other flags, to speed up checking and
shorten commands.
---
ChangeLog | 12 +++
build-aux/gcc-warning.spec | 144 ++++++++++++++++++++++++--
m4/manywarnings.m4 | 206 ++++++-------------------------------
3 files changed, 181 insertions(+), 181 deletions(-)
diff --git a/ChangeLog b/ChangeLog
index edb63b348..6115ab946 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,17 @@
2020-07-01 Paul Eggert <eggert@cs.ucla.edu>
+ manywarnings: improve port to GCC 10.1
+ * build-aux/gcc-warning.spec: Also list warnings that are default
+ or are enabled by already-given flags. This lets us speed up
+ checking for attributes, and makes the generated compilation
+ commands shorter. Add -Wanalyzer-too-complex (too much noise).
+ * m4/manywarnings.m4 (gl_MANYWARN_COMPLEMENT)
+ (gl_MANYWARN_ALL_GCC): Use gl_AS_VAR_APPEND to append
+ to shell variables that may have long values.
+ (gl_MANYWARN_ALL_GCC): Omit flags that are default or are
+ consequences of other flags, to speed up checking and
+ shorten commands.
+
tests: pacify gcc -fanalyzer on zerosize_ptr
* tests/test-memcasecmp.c (main):
* tests/test-memchr.c (main):
diff --git a/build-aux/gcc-warning.spec b/build-aux/gcc-warning.spec
index 89a0bc734..6c0602be4 100644
--- a/build-aux/gcc-warning.spec
+++ b/build-aux/gcc-warning.spec
@@ -1,24 +1,51 @@
# options to filter out, and why
--all-warnings alias for -Wall
--extra-warnings alias for -Wextra
+-W alias for -Wextra
-Wabi this is now a no-op
-Wabi-tag c++
-Wabi= c++
+-Wabsolute-value enabled by -Wextra
+-Waddress enabled by -Wall
+-Waddress-of-packed-member default
-Waggregate-return obsolescent
+-Waggressive-loop-optimizations default
-Waliasing fortran
-Walign-commons fortran
-Waligned-new=[none|global|all] c++
--Walloc-size-larger-than=<bytes> handled specially by gl_MANYWARN_ALL_GCC
+-Walloc-size-larger-than=<bytes> defaults to PTRDIFF_MAX
-Walloc-zero Gnulib fixes this problem
-Walloca we like alloca in small doses
-Walloca-larger-than=<number> FIXME: choose something sane?
-Wampersand fortran
+-Wanalyzer-double-fclose enabled by -fanalyzer
+-Wanalyzer-double-free enabled by -fanalyzer
+-Wanalyzer-exposure-through-output-file enabled by -fanalyzer
+-Wanalyzer-file-leak enabled by -fanalyzer
+-Wanalyzer-free-of-non-heap enabled by -fanalyzer
+-Wanalyzer-malloc-leak enabled by -fanalyzer
+-Wanalyzer-null-argument enabled by -fanalyzer
+-Wanalyzer-null-dereference enabled by -fanalyzer
+-Wanalyzer-possible-null-argument enabled by -fanalyzer
+-Wanalyzer-possible-null-dereference enabled by -fanalyzer
+-Wanalyzer-stale-setjmp-buffer implied by -fanalyzer
+-Wanalyzer-tainted-array-index FIXME maybe? too much noise
+-Wanalyzer-too-complex enabled by -fanalyzer
+-Wanalyzer-unsafe-call-within-signal-handler enabled by -fanalyzer
+-Wanalyzer-use-after-free enabled by -fanalyzer
+-Wanalyzer-use-of-pointer-in-stale-stack-frame enabled by -fanalyzer
-Warray-bounds covered by -Warray-bounds=
-Warray-bounds=<0,2> handled specially by gl_MANYWARN_ALL_GCC
-Warray-temporaries fortran
-Wassign-intercept objc/objc++
-Wattribute-alias covered by -Wattribute-alias=2
-Wattribute-alias=<0,2> handled specially by gl_MANYWARN_ALL_GCC
+-Wattribute-warning default
+-Wattributes default
+-Wbool-compare enabled by -Wall
+-Wbool-operation enabled by -Wall
+-Wbuiltin-declaration-mismatch default
+-Wbuiltin-macro-redefined default
-Wc++-compat only useful for code meant to be compiled by a C++ compiler
-Wc++0x-compat c++
-Wc++11-compat c++
@@ -31,63 +58,120 @@
-Wc11-c2x-compat c compatibility
-Wc90-c99-compat c compatibility
-Wc99-c11-compat c compatibility
+-Wcannot-profile default
+-Wcast-align enabled by -Wcast-align=strict
+-Wcast-function-type enabled by -Wextra
-Wcast-qual FIXME maybe? too much noise; encourages bad changes
-Wcast-result D
-Wcatch-value c++
-Wcatch-value=<0,3> c++
+-Wchar-subscripts enabled by -Wall
-Wcharacter-truncation fortran
-Wchkp deprecated
-Wclass-conversion c++ and objc++
-Wclass-memaccess c++
+-Wclobbered enabled by -Wextra
-Wcomma-subscript c++ and objc++
+-Wcomment enabled by -Wall
+-Wcomments alias for -Wcomment
-Wcompare-reals fortran
-Wconditionally-supported c++ and objc++
-Wconversion FIXME maybe? too much noise; encourages bad changes
-Wconversion-extra fortran
-Wconversion-null c++ and objc++
+-Wcoverage-mismatch default
+-Wcpp default
-Wctor-dtor-privacy c++
+-Wdangling-else enabled by -Wparentheses
-Wdeclaration-after-statement FIXME: do not want. others may
-Wdelete-incomplete c++ and objc++
-Wdelete-non-virtual-dtor c++
+-Wdeprecated default
-Wdeprecated-copy c++ and objc++
-Wdeprecated-copy-dtor c++ and objc++
+-Wdeprecated-declarations default
+-Wdesignated-init default
+-Wdiscarded-array-qualifiers default
+-Wdiscarded-qualifiers default
-Wdo-subscript fortran
+-Wdiv-by-zero default
+-Wduplicate-decl-specifier enabled by -Wall
-Weffc++ c++
+-Wempty-body enabled by -Wextra
+-Wenum-compare enabled by -Wall
+-Wenum-conversion enabled by -Wextra
+-Wendif-labels default
-Werror-implicit-function-declaration deprecated
+-Wexpansion-to-defined enabled by -Wextra
-Wextra-semi c++
-Wfloat-conversion FIXME maybe? borderline. some will want this
-Wfloat-equal FIXME maybe? borderline. some will want this
-Wformat covered by -Wformat=2
+-Wformat-contains-nul default
+-Wformat-diag enabled by -Wformat=2
+-Wformat-extra-args enabled by -Wformat=2
+-Wformat-nonliteral enabled by -Wformat=2
-Wformat-overflow<0,2> gcc --help=warnings artifact
-Wformat-overflow=<0,2> handled specially by gl_MANYWARN_ALL_GCC
+-Wformat-security enabled by -Wformat=2
-Wformat-truncation covered by -Wformat-truncation=2
-Wformat-truncation=<0,2> handled specially by gl_MANYWARN_ALL_GCC
+-Wformat-y2k enabled by -Wformat=2
+-Wformat-zero-length enabled by -Wformat=2
-Wformat=<0,2> gcc --help=warnings artifact
+-Wframe-address enabled by -Wall
-Wframe-larger-than=<byte-size> FIXME: choose something sane?
+-Wfree-nonheap-object default
-Wfunction-elimination fortran
+-Whsa default
+-Wif-not-aligned default
+-Wignored-attributes default
+-Wignored-qualifiers enabled by -Wextra
+-Wimplicit enabled by -Wall
-Wimplicit-fallthrough covered by -Wimplicit-fallthrough=2
-Wimplicit-fallthrough=<0,5> handled specially by gl_MANYWARN_ALL_GCC
+-Wimplicit-function-declaration enabled by -Wimplicit
+-Wimplicit-int enabled by -Wimplicit
-Wimplicit-interface fortran
-Wimplicit-procedure fortran
-Winaccessible-base c++ and objc++
+-Wincompatible-pointer-types default
-Winherited-variadic-ctor c++
-Winit-list-lifetime c++ and objc++
+-Wint-conversion default
+-Wint-in-bool-context enabled by -Wall
+-Wint-to-pointer-cast default
-Winteger-division fortran
-Wintrinsic-shadow fortran
-Wintrinsics-std fortran
+-Winvalid-memory-model default
-Winvalid-offsetof c++ and objc++
-Wjump-misses-init only useful for code meant to be compiled by a C++ compiler
-Wlarger-than- gcc --help=warnings artifact
-Wlarger-than=<byte-size> FIXME: choose something sane?
-Wline-truncation fortran
-Wliteral-suffix c++ and objc++
+-Wlogical-not-parentheses enabled by -Wall
-Wlong-long obsolescent
-Wlto-type-mismatch c++ and objc++
+-Wmain enabled by -Wall
+-Wmaybe-uninitialized enabled by -Wall or -Wextra
+-Wmemset-elt-size enabled by -Wall
+-Wmemset-transposed-args enabled by -Wall
+-Wmisleading-indentation enabled by -Wall
-Wmismatched-tags c++ and objc++
+-Wmissing-attributes enabled by -Wall
+-Wmissing-braces enabled by -Wall
+-Wmissing-field-initializers enabled by -Wextra
-Wmissing-format-attribute obsolescent
-Wmissing-noreturn obsolescent
+-Wmissing-parameter-type enabled by -Wextra
+-Wmissing-profile default
+-Wmultichar default
-Wmultiple-inheritance c++ and objc++
+-Wmultistatement-macros enabled by -Wall
-Wnamespaces c++
+-Wnarrowing enabled by -Wall
-Wno-alloc-size-larger-than see -Walloc-size-larger-than
-Wno-alloca-larger-than see -Walloca-larger-than
-Wno-frame-larger-than see -Wframe-larger-than
@@ -98,21 +182,35 @@
-Wnoexcept-type c++
-Wnon-template-friend c++
-Wnon-virtual-dtor c++
--Wnormalized covered by -Wnormalized=
--Wnormalized=[none|id|nfc|nfkc] handled specially by gl_MANYWARN_ALL_GCC
+-Wnonnull enabled by -Wall or -Wformat
+-Wnonnull-compare enabled by -Wall
+-Wnormalized default
+-Wnormalized=[none|id|nfc|nfkc] defaults to nfc
+-Wodr default
-Wold-style-cast c++ and objc++
+-Wold-style-declaration enabled by -Wextra
+-Woverflow default
-Woverloaded-virtual c++
+-Woverride-init enabled by -Wextra
-Woverride-init-side-effects c++ and objc++
-Woverwrite-recursive fortran
+-Wpacked-bitfield-compat default
+-Wpacked-not-aligned enabled by -Wall
-Wpadded FIXME maybe? warns about "stabil" member in /usr/include/bits/timex.h
+-Wparentheses enabled by -Wall
-Wpedantic FIXME: too strict?
-Wpessimizing-move c++ and objc++
-Wplacement-new c++
-Wplacement-new=<0,2> c++
-Wpmf-conversions c++ and objc++
+-Wpointer-compare default
+-Wpointer-sign enabled by -Wall
+-Wpointer-to-int-cast default
+-Wpragmas default
-Wprio-ctor-dtor c++
-Wproperty-assign-default objc++
-Wprotocol objc++
+-Wpsabi default
-Wreal-q-constant fortran
-Wrealloc-lhs fortran
-Wrealloc-lhs-all fortran
@@ -121,53 +219,87 @@
-Wredundant-tags c++ and objc++
-Wregister c++ and objc++
-Wreorder c++ and objc++
+-Wrestrict enabled by -Wall
+-Wreturn-local-addr default
+-Wreturn-type enabled by -Wall
+-Wscalar-storage-order default
-Wselector objc and objc++
+-Wsequence-point enabled by -Wall
-Wshadow-compatible-local covered by -Wshadow
-Wshadow-ivar objc
-Wshadow-local covered by -Wshadow
-Wshadow=compatible-local covered by -Wshadow
-Wshadow=global covered by -Wshadow
-Wshadow=local covered by -Wshadow
+-Wshift-count-negative default
+-Wshift-count-overflow default
+-Wshift-negative-value enabled by -Wextra
-Wshift-overflow covered by -Wshift-overflow=2
-Wshift-overflow=<0,2> gcc --help=warnings artifact
-Wsign-compare FIXME maybe? borderline. some will want this
-Wsign-conversion FIXME maybe? borderline. some will want this
-Wsign-promo c++ and objc++
-Wsized-deallocation c++ and objc++
+-Wsizeof-array-argument default
+-Wsizeof-pointer-div enabled by -Wall
+-Wsizeof-pointer-memaccess enabled by -Wall
-Wstack-usage=<byte-size> FIXME: choose something sane?
+-Wstrict-aliasing enabled by -Wall
-Wstrict-aliasing=<0,3> FIXME: choose something sane?
-Wstrict-null-sentinel c++ and objc++
-Wstrict-overflow=<0,5> FIXME: choose something sane?
-Wstrict-selector-match objc and objc++
--Wstringop-overflow covered by -Wstringop-overflow=
--Wstringop-overflow=<0,4> handled specially by gl_MANYWARN_ALL_GCC
+-Wstring-compare enabled by -Wextra
+-Wstringop-overflow covered by -Wstringop-overflow=2
+-Wstringop-overflow=<0,4> defaults to 2
+-Wstringop-truncation default
-Wsubobject-linkage c++ and objc++
-Wsuggest-override c++ and objc++
-Wsurprising fortran
+-Wswitch enabled by -Wall
+-Wswitch-bool default
-Wswitch-default https://lists.gnu.org/r/bug-gnulib/2018-05/msg00179.html
-Wswitch-enum FIXME maybe? borderline. some will want this
+-Wswitch-outside-range default
+-Wswitch-unreachable default
-Wsynth deprecated
-Wtabs fortran
-Wtarget-lifetime fortran
+-Wtautological-compare enabled by -Wall
-Wtemplates c++ and objc++
-Wterminate c++ and objc++
-Wtraditional obsolescent
-Wtraditional-conversion obsolescent
+-Wtrigraphs enabled by -Wall
+-Wtype-limits enabled by -Wextra
-Wundeclared-selector objc and objc++
-Wundef FIXME maybe? too many false positives
-Wundefined-do-loop fortran
-Wunderflow fortran
-Wunreachable-code obsolescent no-op
-Wunsuffixed-float-constants triggers warning in gnulib's timespec.h
--Wunused-const-variable covered by -Wunusec-const-variable=2
+-Wunused enabled by -Wall
+-Wunused-but-set-parameter enabled by -Wunused
+-Wunused-but-set-variable enabled by -Wunused
+-Wunused-const-variable covered by -Wunused-const-variable=2
-Wunused-const-variable=<0,2> gcc --help=warnings artifact
-Wunused-dummy-argument fortran
+-Wunused-function enabled by -Wunused
+-Wunused-label enabled by -Wunused
+-Wunused-local-typedefs enabled by -Wunused
+-Wunused-parameter enabled by -Wunused
+-Wunused-result enabled by -Wunused
+-Wunused-value enabled by -Wunused
+-Wunused-variable enabled by -Wunused
-Wuse-without-only fortran
-Wuseless-cast c++ and objc++
+-Wvarargs default
-Wvirtual-inheritance c++
-Wvirtual-move-assign c++
-Wvla-larger-than=<number> handled specially by gl_MANYWARN_ALL_GCC
-Wvolatile c++ and objc++
+-Wvolatile-register-var enabled by -Wall
-Wzero-as-null-pointer-constant c++ and objc++
+-Wzero-length-bounds enabled by -Wall
-Wzerotrip fortran
-frequire-return-statement go
diff --git a/m4/manywarnings.m4 b/m4/manywarnings.m4
index 719bafb29..d18da048d 100644
--- a/m4/manywarnings.m4
+++ b/m4/manywarnings.m4
@@ -1,4 +1,4 @@
-# manywarnings.m4 serial 19
+# manywarnings.m4 serial 20
dnl Copyright (C) 2008-2020 Free Software Foundation, Inc.
dnl This file is free software; the Free Software Foundation
dnl gives unlimited permission to copy and/or distribute it,
@@ -21,7 +21,7 @@ AC_DEFUN([gl_MANYWARN_COMPLEMENT],
*" $gl_warn_item "*)
;;
*)
- gl_warn_set="$gl_warn_set $gl_warn_item"
+ gl_AS_VAR_APPEND([gl_warn_set], [" $gl_warn_item"])
;;
esac
done
@@ -49,12 +49,12 @@ m4_defun([gl_MANYWARN_ALL_GCC(C)],
AC_REQUIRE([AC_PROG_CC])
if test -n "$GCC"; then
- dnl Check if -W -Werror -Wno-missing-field-initializers is supported
+ dnl Check if -Wextra -Werror -Wno-missing-field-initializers is supported
dnl with the current $CC $CFLAGS $CPPFLAGS.
AC_CACHE_CHECK([whether -Wno-missing-field-initializers is supported],
[gl_cv_cc_nomfi_supported],
[gl_save_CFLAGS="$CFLAGS"
- CFLAGS="$CFLAGS -W -Werror -Wno-missing-field-initializers"
+ CFLAGS="$CFLAGS -Wextra -Werror -Wno-missing-field-initializers"
AC_COMPILE_IFELSE(
[AC_LANG_PROGRAM([[]], [[]])],
[gl_cv_cc_nomfi_supported=yes],
@@ -68,7 +68,7 @@ m4_defun([gl_MANYWARN_ALL_GCC(C)],
AC_CACHE_CHECK([whether -Wno-missing-field-initializers is needed],
[gl_cv_cc_nomfi_needed],
[gl_save_CFLAGS="$CFLAGS"
- CFLAGS="$CFLAGS -W -Werror"
+ CFLAGS="$CFLAGS -Wextra -Werror"
AC_COMPILE_IFELSE(
[AC_LANG_PROGRAM(
[[int f (void)
@@ -105,153 +105,41 @@ m4_defun([gl_MANYWARN_ALL_GCC(C)],
# To compare this list to your installed GCC's, run this Bash command:
#
# comm -3 \
- # <((sed -n 's/^ *\(-[^ 0-9][^ ]*\) .*/\1/p' manywarnings.m4; \
+ # <((sed -n 's/^ *\(-[^ 0-9][^ ]*\).*/\1/p' manywarnings.m4; \
# awk '/^[^#]/ {print $1}' ../build-aux/gcc-warning.spec) | sort) \
# <(LC_ALL=C gcc --help=warnings | sed -n 's/^ \(-[^ ]*\) .*/\1/p' | sort)
- gl_manywarn_set=
- for gl_manywarn_item in -fno-common \
- -W \
- -Wabsolute-value \
- -Waddress \
- -Waddress-of-packed-member \
- -Waggressive-loop-optimizations \
+ $1=
+ for gl_manywarn_item in -fanalyzer -fno-common \
-Wall \
- -Wanalyzer-double-fclose \
- -Wanalyzer-double-free \
- -Wanalyzer-exposure-through-output-file \
- -Wanalyzer-file-leak \
- -Wanalyzer-free-of-non-heap \
- -Wanalyzer-malloc-leak \
- -Wanalyzer-null-argument \
- -Wanalyzer-null-dereference \
- -Wanalyzer-possible-null-argument \
- -Wanalyzer-possible-null-dereference \
- -Wanalyzer-stale-setjmp-buffer \
- -Wanalyzer-tainted-array-index \
- -Wanalyzer-too-complex \
- -Wanalyzer-unsafe-call-within-signal-handler \
- -Wanalyzer-use-after-free \
- -Wanalyzer-use-of-pointer-in-stale-stack-frame \
-Warith-conversion \
- -Wattribute-warning \
- -Wattributes \
-Wbad-function-cast \
- -Wbool-compare \
- -Wbool-operation \
- -Wbuiltin-declaration-mismatch \
- -Wbuiltin-macro-redefined \
- -Wcannot-profile \
- -Wcast-align \
-Wcast-align=strict \
- -Wcast-function-type \
- -Wchar-subscripts \
- -Wclobbered \
- -Wcomment \
- -Wcomments \
- -Wcoverage-mismatch \
- -Wcpp \
- -Wdangling-else \
-Wdate-time \
- -Wdeprecated \
- -Wdeprecated-declarations \
- -Wdesignated-init \
-Wdisabled-optimization \
- -Wdiscarded-array-qualifiers \
- -Wdiscarded-qualifiers \
- -Wdiv-by-zero \
-Wdouble-promotion \
-Wduplicated-branches \
-Wduplicated-cond \
- -Wduplicate-decl-specifier \
- -Wempty-body \
- -Wendif-labels \
- -Wenum-compare \
- -Wenum-conversion \
- -Wexpansion-to-defined \
-Wextra \
- -Wformat-contains-nul \
- -Wformat-diag \
- -Wformat-extra-args \
- -Wformat-nonliteral \
- -Wformat-security \
-Wformat-signedness \
- -Wformat-y2k \
- -Wformat-zero-length \
- -Wframe-address \
- -Wfree-nonheap-object \
- -Whsa \
- -Wif-not-aligned \
- -Wignored-attributes \
- -Wignored-qualifiers \
- -Wimplicit \
- -Wimplicit-function-declaration \
- -Wimplicit-int \
- -Wincompatible-pointer-types \
-Winit-self \
-Winline \
- -Wint-conversion \
- -Wint-in-bool-context \
- -Wint-to-pointer-cast \
- -Winvalid-memory-model \
-Winvalid-pch \
- -Wlogical-not-parentheses \
-Wlogical-op \
- -Wmain \
- -Wmaybe-uninitialized \
- -Wmemset-elt-size \
- -Wmemset-transposed-args \
- -Wmisleading-indentation \
- -Wmissing-attributes \
- -Wmissing-braces \
-Wmissing-declarations \
- -Wmissing-field-initializers \
-Wmissing-include-dirs \
- -Wmissing-parameter-type \
- -Wmissing-profile \
-Wmissing-prototypes \
- -Wmultichar \
- -Wmultistatement-macros \
- -Wnarrowing \
-Wnested-externs \
- -Wnonnull \
- -Wnonnull-compare \
-Wnull-dereference \
- -Wodr \
- -Wold-style-declaration \
-Wold-style-definition \
-Wopenmp-simd \
- -Woverflow \
-Woverlength-strings \
- -Woverride-init \
-Wpacked \
- -Wpacked-bitfield-compat \
- -Wpacked-not-aligned \
- -Wparentheses \
-Wpointer-arith \
- -Wpointer-compare \
- -Wpointer-sign \
- -Wpointer-to-int-cast \
- -Wpragmas \
- -Wpsabi \
- -Wrestrict \
- -Wreturn-local-addr \
- -Wreturn-type \
- -Wscalar-storage-order \
- -Wsequence-point \
-Wshadow \
- -Wshift-count-negative \
- -Wshift-count-overflow \
- -Wshift-negative-value \
- -Wsizeof-array-argument \
- -Wsizeof-pointer-div \
- -Wsizeof-pointer-memaccess \
-Wstack-protector \
- -Wstrict-aliasing \
-Wstrict-overflow \
-Wstrict-prototypes \
- -Wstring-compare \
- -Wstringop-truncation \
-Wsuggest-attribute=cold \
-Wsuggest-attribute=const \
-Wsuggest-attribute=format \
@@ -260,95 +148,63 @@ m4_defun([gl_MANYWARN_ALL_GCC(C)],
-Wsuggest-attribute=pure \
-Wsuggest-final-methods \
-Wsuggest-final-types \
- -Wswitch \
- -Wswitch-bool \
- -Wswitch-outside-range \
- -Wswitch-unreachable \
-Wsync-nand \
-Wsystem-headers \
- -Wtautological-compare \
-Wtrampolines \
- -Wtrigraphs \
- -Wtype-limits \
-Wuninitialized \
-Wunknown-pragmas \
-Wunsafe-loop-optimizations \
- -Wunused \
- -Wunused-but-set-parameter \
- -Wunused-but-set-variable \
- -Wunused-function \
- -Wunused-label \
- -Wunused-local-typedefs \
-Wunused-macros \
- -Wunused-parameter \
- -Wunused-result \
- -Wunused-value \
- -Wunused-variable \
- -Wvarargs \
-Wvariadic-macros \
-Wvector-operation-performance \
-Wvla \
- -Wvolatile-register-var \
-Wwrite-strings \
- -Wzero-length-bounds \
\
; do
- gl_manywarn_set="$gl_manywarn_set $gl_manywarn_item"
+ gl_AS_VAR_APPEND([$1], [" $gl_manywarn_item"])
done
# gcc --help=warnings outputs an unusual form for these options; list
# them here so that the above 'comm' command doesn't report a false match.
- # Would prefer "min (PTRDIFF_MAX, SIZE_MAX)", but it must be a literal.
- # Also, AC_COMPUTE_INT requires it to fit in a long; it is 2**63 on
- # the only platforms where it does not fit in a long, so make that
- # a special case.
- AC_MSG_CHECKING([max safe object size])
- AC_COMPUTE_INT([gl_alloc_max],
- [LONG_MAX < (PTRDIFF_MAX < (size_t) -1 ? PTRDIFF_MAX : (size_t) -1)
- ? -1
- : PTRDIFF_MAX < (size_t) -1 ? (long) PTRDIFF_MAX : (long) (size_t) -1],
- [[#include <limits.h>
- #include <stddef.h>
- #include <stdint.h>
- ]],
- [gl_alloc_max=2147483647])
- case $gl_alloc_max in
- -1) gl_alloc_max=9223372036854775807;;
- esac
- AC_MSG_RESULT([$gl_alloc_max])
- gl_manywarn_set="$gl_manywarn_set -Walloc-size-larger-than=$gl_alloc_max"
- gl_manywarn_set="$gl_manywarn_set -Warray-bounds=2"
- gl_manywarn_set="$gl_manywarn_set -Wattribute-alias=2"
- gl_manywarn_set="$gl_manywarn_set -Wformat-overflow=2"
- gl_manywarn_set="$gl_manywarn_set -Wformat-truncation=2"
- gl_manywarn_set="$gl_manywarn_set -Wimplicit-fallthrough=5"
- gl_manywarn_set="$gl_manywarn_set -Wnormalized=nfc"
- gl_manywarn_set="$gl_manywarn_set -Wshift-overflow=2"
- gl_manywarn_set="$gl_manywarn_set -Wstringop-overflow=2"
- gl_manywarn_set="$gl_manywarn_set -Wunused-const-variable=2"
- gl_manywarn_set="$gl_manywarn_set -Wvla-larger-than=4031"
+ gl_AS_VAR_APPEND([$1], [' -Warray-bounds=2'])
+ gl_AS_VAR_APPEND([$1], [' -Wattribute-alias=2'])
+ gl_AS_VAR_APPEND([$1], [' -Wformat-overflow=2'])
+ gl_AS_VAR_APPEND([$1], [' -Wformat=2'])
+ gl_AS_VAR_APPEND([$1], [' -Wformat-truncation=2'])
+ gl_AS_VAR_APPEND([$1], [' -Wimplicit-fallthrough=5'])
+ gl_AS_VAR_APPEND([$1], [' -Wshift-overflow=2'])
+ gl_AS_VAR_APPEND([$1], [' -Wunused-const-variable=2'])
+ gl_AS_VAR_APPEND([$1], [' -Wvla-larger-than=4031'])
# These are needed for older GCC versions.
if test -n "$GCC"; then
case `($CC --version) 2>/dev/null` in
'gcc (GCC) '[[0-3]].* | \
'gcc (GCC) '4.[[0-7]].*)
- gl_manywarn_set="$gl_manywarn_set -fdiagnostics-show-option"
- gl_manywarn_set="$gl_manywarn_set -funit-at-a-time"
+ gl_AS_VAR_APPEND([$1], [' -fdiagnostics-show-option'])
+ gl_AS_VAR_APPEND([$1], [' -funit-at-a-time'])
;;
esac
fi
# Disable specific options as needed.
if test "$gl_cv_cc_nomfi_needed" = yes; then
- gl_manywarn_set="$gl_manywarn_set -Wno-missing-field-initializers"
+ gl_AS_VAR_APPEND([$1], [' -Wno-missing-field-initializers'])
fi
if test "$gl_cv_cc_uninitialized_supported" = no; then
- gl_manywarn_set="$gl_manywarn_set -Wno-uninitialized"
+ gl_AS_VAR_APPEND([$1], [' -Wno-uninitialized'])
fi
- $1=$gl_manywarn_set
+ # Some warnings have too many false alarms in GCC 10.1.
+ # https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93695
+ gl_AS_VAR_APPEND([$1], [' -Wno-analyzer-double-free'])
+ # https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94458
+ gl_AS_VAR_APPEND([$1], [' -Wno-analyzer-malloc-leak'])
+ # https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94851
+ gl_AS_VAR_APPEND([$1], [' -Wno-analyzer-null-dereference'])
+ # https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95758
+ gl_AS_VAR_APPEND([$1], [' -Wno-analyzer-use-after-free'])
AC_LANG_POP([C])
])
--
2.17.1
^ permalink raw reply related [flat|nested] 7+ messages in thread
* Re: coreutils and GCC -fanalyzer
2020-07-02 0:26 ` Paul Eggert
@ 2020-07-10 21:21 ` Pádraig Brady
2020-07-10 21:37 ` Jeffrey Walton
2020-07-11 0:58 ` Paul Eggert
0 siblings, 2 replies; 7+ messages in thread
From: Pádraig Brady @ 2020-07-10 21:21 UTC (permalink / raw)
To: Paul Eggert; +Cc: Gnulib bugs, coreutils
On 02/07/2020 01:26, Paul Eggert wrote:
> On 5/23/20 9:08 AM, Paul Eggert wrote:
>
>> So I am thinking of killing two
>> stones by doing the following.
>>
>> 1. Test for -fanalyzer, -Wall, -Wextra.
>>
>> 2. Test for flags not automatically enabled by -fanalyzer, -Wall, -Wextra but
>> flags that we want anyway.
>>
>> 3. Test for flags automatically enabled by -fanalyzer, -Wall, -Wextra that are
>> also flags that we don't want.
>
> I did that in Gnulib by installing the attached patch. This could greatly
> increase compile times due to the -fanalyzer option, so let's keep an eye out
> for that.
s/could/does/.
compile time has gone from seconds to minutes for coreutils at least.
I'd be inclined to not enable -fanalyzer by default.
At least not until it matures more.
-fanalyzer didn't find any actual issues in coreutils,
and currently bails on most files (after taking lots of time about it),
as indicated with the -Wanalyzer-too-complex option.
cheers,
Pádraig
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: coreutils and GCC -fanalyzer
2020-07-10 21:21 ` Pádraig Brady
@ 2020-07-10 21:37 ` Jeffrey Walton
2020-07-11 0:58 ` Paul Eggert
1 sibling, 0 replies; 7+ messages in thread
From: Jeffrey Walton @ 2020-07-10 21:37 UTC (permalink / raw)
To: bug-gnulib, coreutils
On Fri, Jul 10, 2020 at 5:22 PM Pádraig Brady <P@draigbrady.com> wrote:
>
> On 02/07/2020 01:26, Paul Eggert wrote:
> > On 5/23/20 9:08 AM, Paul Eggert wrote:
> >
> >> So I am thinking of killing two
> >> stones by doing the following.
> >>
> >> 1. Test for -fanalyzer, -Wall, -Wextra.
> >>
> >> 2. Test for flags not automatically enabled by -fanalyzer, -Wall, -Wextra but
> >> flags that we want anyway.
> >>
> >> 3. Test for flags automatically enabled by -fanalyzer, -Wall, -Wextra that are
> >> also flags that we don't want.
> >
> > I did that in Gnulib by installing the attached patch. This could greatly
> > increase compile times due to the -fanalyzer option, so let's keep an eye out
> > for that.
>
> s/could/does/.
> compile time has gone from seconds to minutes for coreutils at least.
>
> I'd be inclined to not enable -fanalyzer by default.
> At least not until it matures more.
> -fanalyzer didn't find any actual issues in coreutils,
> and currently bails on most files (after taking lots of time about it),
> as indicated with the -Wanalyzer-too-complex option.
Just my 2 cents, but leave analyzer options to testing and QA folks.
Make the option available for convenience, but leave it off by
default.
The testers and QA folks will perform the enhanced testing on
occasion, like during a CI build or manually before a release. The
remainder of the time, regular users get a regular build.
Jeff
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: coreutils and GCC -fanalyzer
2020-07-10 21:21 ` Pádraig Brady
2020-07-10 21:37 ` Jeffrey Walton
@ 2020-07-11 0:58 ` Paul Eggert
2020-07-11 14:26 ` Pádraig Brady
1 sibling, 1 reply; 7+ messages in thread
From: Paul Eggert @ 2020-07-11 0:58 UTC (permalink / raw)
To: Pádraig Brady; +Cc: Gnulib bugs, coreutils
[-- Attachment #1: Type: text/plain, Size: 620 bytes --]
On 7/10/20 2:21 PM, Pádraig Brady wrote:
>
> I'd be inclined to not enable -fanalyzer by default.
> At least not until it matures more.
> -fanalyzer didn't find any actual issues in coreutils,
Yes, the only bug I found related to coreutils:
https://gmplib.org/list-archives/gmp-bugs/2020-July/004844.html
is a bug in library code that coreutils never triggers.
> I'd be inclined to not enable -fanalyzer by default.
I installed the attached patch into coreutils; will that do? The basic idea is
that you get -fanalyzer only if you configure with the new
--enable-gcc-warnings=expensive option.
[-- Attachment #2: 0001-build-be-less-aggressive-about-fanalyzer.patch --]
[-- Type: text/x-patch, Size: 2658 bytes --]
From 6b6f0f54c0eca06345b7741e5fd37b4500675286 Mon Sep 17 00:00:00 2001
From: Paul Eggert <eggert@cs.ucla.edu>
Date: Fri, 10 Jul 2020 15:54:51 -0700
Subject: [PATCH] build: be less aggressive about -fanalyzer
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
* configure.ac: Don’t enable -fanalyzer unless configured with the
new --enable-gcc-warnings=expensive option. See thread at:
https://lists.gnu.org/r/coreutils/2020-07/msg00011.html
---
configure.ac | 22 ++++++++++++++++------
1 file changed, 16 insertions(+), 6 deletions(-)
diff --git a/configure.ac b/configure.ac
index 0afbff627..69d4e7d7f 100644
--- a/configure.ac
+++ b/configure.ac
@@ -89,10 +89,14 @@ AC_DEFUN([gl_GCC_VERSION_IFELSE],
)
AC_ARG_ENABLE([gcc-warnings],
- [AS_HELP_STRING([--enable-gcc-warnings],
- [turn on many GCC warnings (for developers; best with GNU make)])],
+ [AS_HELP_STRING([--enable-gcc-warnings@<:@=TYPE@:>@],
+ [control generation of GCC warnings. The TYPE 'no' disables
+ warnings (default for non-developer builds); 'yes' generates
+ cheap warnings if available (default for developer builds);
+ 'expensive' in addition generates expensive-to-compute warnings
+ if available.])],
[case $enableval in
- yes|no) ;;
+ no|yes|expensive) ;;
*) AC_MSG_ERROR([bad value $enableval for gcc-warnings option]) ;;
esac
gl_gcc_warnings=$enableval],
@@ -108,12 +112,18 @@ AC_ARG_ENABLE([gcc-warnings],
&& gl_gcc_warnings=yes])]
)
-if test "$gl_gcc_warnings" = yes; then
+if test $gl_gcc_warnings != no; then
gl_WARN_ADD([-Werror], [WERROR_CFLAGS])
AC_SUBST([WERROR_CFLAGS])
- nw=
+ ew=
+ AS_IF([test $gl_gcc_warnings != expensive],
+ [# -fanalyzer and related options slow GCC considerably.
+ ew="$ew -fanalyzer -Wno-analyzer-double-free -Wno-analyzer-malloc-leak"
+ ew="$ew -Wno-analyzer-null-dereference -Wno-analyzer-use-after-free"])
+
# This, $nw, is the list of warnings we disable.
+ nw=$ew
nw="$nw -Wdeclaration-after-statement" # too useful to forbid
nw="$nw -Waggregate-return" # anachronistic
nw="$nw -Wlong-long" # C90 is anachronistic (lib/gethrxtime.h)
@@ -190,7 +200,7 @@ if test "$gl_gcc_warnings" = yes; then
# We use a slightly smaller set of warning options for lib/.
# Remove the following and save the result in GNULIB_WARN_CFLAGS.
- nw=
+ nw=$ew
nw="$nw -Wduplicated-branches" # Too many false alarms
nw="$nw -Wformat-truncation=2"
nw="$nw -Wstrict-overflow"
--
2.17.1
^ permalink raw reply related [flat|nested] 7+ messages in thread
* Re: coreutils and GCC -fanalyzer
2020-07-11 0:58 ` Paul Eggert
@ 2020-07-11 14:26 ` Pádraig Brady
0 siblings, 0 replies; 7+ messages in thread
From: Pádraig Brady @ 2020-07-11 14:26 UTC (permalink / raw)
To: Paul Eggert; +Cc: Gnulib bugs, coreutils
On 11/07/2020 01:58, Paul Eggert wrote:
> On 7/10/20 2:21 PM, Pádraig Brady wrote:
>>
>> I'd be inclined to not enable -fanalyzer by default.
>> At least not until it matures more.
>> -fanalyzer didn't find any actual issues in coreutils,
>
> Yes, the only bug I found related to coreutils:
>
> https://gmplib.org/list-archives/gmp-bugs/2020-July/004844.html
>
> is a bug in library code that coreutils never triggers.
>
>> I'd be inclined to not enable -fanalyzer by default.
>
> I installed the attached patch into coreutils; will that do? The basic idea is
> that you get -fanalyzer only if you configure with the new
> --enable-gcc-warnings=expensive option.
That's better, for coreutils at least.
It's good that you included all fanalyzer related options set by gnulib,
as that allows projects to more easily control those themselves.
thanks!
Pádraig
^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2020-07-11 14:26 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
[not found] <edfafbde-9e71-f8a9-600d-23cdfaee52a5@cs.ucla.edu>
[not found] ` <77400f8a-59ca-6d02-e5b1-e01ba0619237@draigBrady.com>
[not found] ` <a554e905-fa73-6618-90d7-ae3cb136e19d@cs.ucla.edu>
[not found] ` <e551aac0-e2fe-920a-d0ea-ca80695283d0@draigBrady.com>
[not found] ` <4aa67590-a0c3-603d-d2b1-f0751a28df29@cs.ucla.edu>
[not found] ` <3a074086-47f7-638f-50a8-cd7ceb46a11a@draigBrady.com>
2020-05-23 16:08 ` coreutils and GCC -fanalyzer Paul Eggert
2020-05-23 16:22 ` Tim Rühsen
2020-07-02 0:26 ` Paul Eggert
2020-07-10 21:21 ` Pádraig Brady
2020-07-10 21:37 ` Jeffrey Walton
2020-07-11 0:58 ` Paul Eggert
2020-07-11 14:26 ` Pádraig Brady
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).