Re: gcc 7.3.1 build - warnings as errors in harfbuzz
Hi Matthias,
I've looked a little at this error and it is quite strange. First of
all, I can't reproduce it with a default gcc 7.3.0 and there doesn't
exist an official version of gcc 7.3.1 (at least I couldn't find it).
Also, I can't see the real error in the objected code. The values the
warning prints ("assuming directive output of 2147488582 bytes" and
"output 2 or more bytes (assuming 2147488583) into a destination of
size 127") seem to be excessively large. Every other reference for
this warning I could google reports much small numbers and I don't see
how a double could potentially use 2147488583 characters in the output
buffer.
Could this be a problem with the compiler you've used? Could you also
ask the people who compiled and patched it?
I did found the GCC bug "[7 Regression] -Wformat-overflow false
positive exceeding INT_MAX in glibc sysdeps/posix/tempname.c" [1]
which seem to resemble to the output you see, but that should have
been fixed already in the 7.0 release.
Regards,
Volker
[1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79275
On Wed, Oct 10, 2018 at 3:14 PM Baesken, Matthias
wrote:
>
> Hello , when compiling jdk/jdk with gcc 7.3.1on linux x86_64 (or
> also on linux ppc64) I run into this build error :
>
>
> /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanager/harfbuzz/hb-common.cc:
> In function 'void hb_variation_to_string(hb_variation_t*, char*, unsigned
> int)':
> /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanager/harfbuzz/hb-common.cc:1066:27:
> error: '%g' directive output between 1 and 18446744073709551615 bytes may
> cause result to exceed 'INT_MAX' [-Werror=format-truncation=]
>len += MAX (0, snprintf (s + len, ARRAY_LENGTH (s) - len, "%g",
> variation->value));
>
> ~^
> /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanager/harfbuzz/hb-common.cc:1066:27:
> note: assuming directive output of 2147488582 bytes
> cc1plus: all warnings being treated as errors
>
> (build is a product - build)
>
> Setting –disable-warnings-as-errors works as a workaround , but of course
> this is not really what we want to do .
>
> Fixing in theharfbuzz sources in OpenJDk might be also not so nice
> because it would clash with imports of new versions of harfbuzz .
> Do you think we could disable the specific warning for the library
> compilation ?
>
> Any other great suggestions ?
>
>
> Thanks, Matthias
Re: gcc 7.3.1 build - warnings as errors in harfbuzz
GCC (since some version older than we support now) ignores disabling of unknown warning labels. This is a feature we depend on for these constructs. Without that it wouldn't be possible to list all the labels like this. (We used to check for this feature in GCC in configure and only use the warning disabling flags if GCC properly ignored unknowns, but since we require a newer version than that now, we no longer need to) /Erik On 2018-10-10 09:10, Baesken, Matthias wrote: Thanks. I am a bit worried that DISABLED_WARNINGS_gcc := format-truncation DISABLED_WARNINGS_CXX_gcc := format-truncation Could bring errors on older gcc versions , any ideas about this ? (will test with gcc 4.8 of course as well ) Best regards, Matthias -Original Message- From: Erik Joelsson Sent: Mittwoch, 10. Oktober 2018 18:08 To: Baesken, Matthias ; 'build- d...@openjdk.java.net' Subject: Re: gcc 7.3.1 build - warnings as errors in harfbuzz I think that's fine. This is the granularity we have. /Erik On 2018-10-10 09:02, Baesken, Matthias wrote: Hi Erik, so I think I could disable the warning here : Awt2dLibraries.gmk - $(eval $(call SetupJdkLibrary, BUILD_LIBFONTMANAGER, \ NAME := fontmanager, \ WARNINGS_AS_ERRORS_xlc := false, \ DISABLED_WARNINGS_gcc := format-truncation sign-compare int-to- pointer-cast \ type-limits missing-field-initializers implicit-fallthrough \ strict-aliasing undef unused-function, \ DISABLED_WARNINGS_CXX_gcc := format-truncation reorder delete- non-virtual-dtor strict-overflow \ maybe-uninitialized, \ ... (add format-truncationfor gcc in the warning-disabling section). However this would disable it also for other versions of gcc where the issue never showed up . Do you think this is fine (and safe for older gcc) ? Best regards, Matthias -Original Message- From: Erik Joelsson Sent: Mittwoch, 10. Oktober 2018 17:33 To: Baesken, Matthias ; 'build- d...@openjdk.java.net' Subject: Re: gcc 7.3.1 build - warnings as errors in harfbuzz In this case, disabling the warning seems like the right thing to do. /Erik On 2018-10-10 06:14, Baesken, Matthias wrote: Hello , when compiling jdk/jdk with gcc 7.3.1on linux x86_64 (or also on linux ppc64) I run into this build error : /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanag er/harfbuzz/hb-common.cc: In function 'void hb_variation_to_string(hb_variation_t*, char*, unsigned int)': /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanag er/harfbuzz/hb-common.cc:1066:27: error: '%g' directive output between 1 and 18446744073709551615 bytes may cause result to exceed 'INT_MAX' [- Werror=format-truncation=] len += MAX (0, snprintf (s + len, ARRAY_LENGTH (s) - len, "%g", variation- value)); ~^ /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanag er/harfbuzz/hb-common.cc:1066:27: note: assuming directive output of 2147488582 bytes cc1plus: all warnings being treated as errors (build is a product - build) Setting –disable-warnings-as-errors works as a workaround , but of course this is not really what we want to do . Fixing in theharfbuzz sources in OpenJDk might be also not so nice because it would clash with imports of new versions of harfbuzz . Do you think we could disable the specific warning for the library compilation ? Any other great suggestions 😉 ? Thanks, Matthias
RE: gcc 7.3.1 build - warnings as errors in harfbuzz
Thanks. I am a bit worried that DISABLED_WARNINGS_gcc := format-truncation DISABLED_WARNINGS_CXX_gcc := format-truncation Could bring errors on older gcc versions , any ideas about this ? (will test with gcc 4.8 of course as well ) Best regards, Matthias > -Original Message- > From: Erik Joelsson > Sent: Mittwoch, 10. Oktober 2018 18:08 > To: Baesken, Matthias ; 'build- > d...@openjdk.java.net' > Subject: Re: gcc 7.3.1 build - warnings as errors in harfbuzz > > I think that's fine. This is the granularity we have. > > /Erik > > > On 2018-10-10 09:02, Baesken, Matthias wrote: > > Hi Erik, so I think I could disable the warning here : > > > > > > Awt2dLibraries.gmk > > - > > > > $(eval $(call SetupJdkLibrary, BUILD_LIBFONTMANAGER, \ > > NAME := fontmanager, \ > > > > WARNINGS_AS_ERRORS_xlc := false, \ > > DISABLED_WARNINGS_gcc := format-truncation sign-compare int-to- > pointer-cast \ > > type-limits missing-field-initializers implicit-fallthrough \ > > strict-aliasing undef unused-function, \ > > DISABLED_WARNINGS_CXX_gcc := format-truncation reorder delete- > non-virtual-dtor strict-overflow \ > > maybe-uninitialized, \ > > ... > > > > (add format-truncationfor gcc in the warning-disabling section). > > However this would disable it also for other versions of gcc where the > issue never showed up . > > > > Do you think this is fine (and safe for older gcc) ? > > > > > > Best regards, Matthias > > > >> -Original Message- > >> From: Erik Joelsson > >> Sent: Mittwoch, 10. Oktober 2018 17:33 > >> To: Baesken, Matthias ; 'build- > >> d...@openjdk.java.net' > >> Subject: Re: gcc 7.3.1 build - warnings as errors in harfbuzz > >> > >> In this case, disabling the warning seems like the right thing to do. > >> > >> /Erik > >> > >> > >> On 2018-10-10 06:14, Baesken, Matthias wrote: > >>> Hello , when compiling jdk/jdk with gcc 7.3.1on linux x86_64 > >>> (or also > >> on linux ppc64) I run into this build error : > >>> > >>> > >> > /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanag > >> er/harfbuzz/hb-common.cc: In function 'void > >> hb_variation_to_string(hb_variation_t*, char*, unsigned int)': > >> > /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanag > >> er/harfbuzz/hb-common.cc:1066:27: error: '%g' directive output between > 1 > >> and 18446744073709551615 bytes may cause result to exceed 'INT_MAX' > [- > >> Werror=format-truncation=] > >>> len += MAX (0, snprintf (s + len, ARRAY_LENGTH (s) - len, "%g", > variation- > >>> value)); > >>> > >> > ~^ > >> > >> > /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanag > >> er/harfbuzz/hb-common.cc:1066:27: note: assuming directive output of > >> 2147488582 bytes > >>> cc1plus: all warnings being treated as errors > >>> > >>> (build is a product - build) > >>> > >>> Setting –disable-warnings-as-errors works as a workaround , but of > >> course this is not really what we want to do . > >>> Fixing in theharfbuzz sources in OpenJDk might be also not so > >>> nice > >> because it would clash with imports of new versions of harfbuzz . > >>> Do you think we could disable the specific warning for the library > >> compilation ? > >>> Any other great suggestions 😉 ? > >>> > >>> > >>> Thanks, Matthias
Re: gcc 7.3.1 build - warnings as errors in harfbuzz
I think that's fine. This is the granularity we have. /Erik On 2018-10-10 09:02, Baesken, Matthias wrote: Hi Erik, so I think I could disable the warning here : Awt2dLibraries.gmk - $(eval $(call SetupJdkLibrary, BUILD_LIBFONTMANAGER, \ NAME := fontmanager, \ WARNINGS_AS_ERRORS_xlc := false, \ DISABLED_WARNINGS_gcc := format-truncation sign-compare int-to-pointer-cast \ type-limits missing-field-initializers implicit-fallthrough \ strict-aliasing undef unused-function, \ DISABLED_WARNINGS_CXX_gcc := format-truncation reorder delete-non-virtual-dtor strict-overflow \ maybe-uninitialized, \ ... (add format-truncationfor gcc in the warning-disabling section). However this would disable it also for other versions of gcc where the issue never showed up . Do you think this is fine (and safe for older gcc) ? Best regards, Matthias -Original Message- From: Erik Joelsson Sent: Mittwoch, 10. Oktober 2018 17:33 To: Baesken, Matthias ; 'build- d...@openjdk.java.net' Subject: Re: gcc 7.3.1 build - warnings as errors in harfbuzz In this case, disabling the warning seems like the right thing to do. /Erik On 2018-10-10 06:14, Baesken, Matthias wrote: Hello , when compiling jdk/jdk with gcc 7.3.1on linux x86_64 (or also on linux ppc64) I run into this build error : /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanag er/harfbuzz/hb-common.cc: In function 'void hb_variation_to_string(hb_variation_t*, char*, unsigned int)': /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanag er/harfbuzz/hb-common.cc:1066:27: error: '%g' directive output between 1 and 18446744073709551615 bytes may cause result to exceed 'INT_MAX' [- Werror=format-truncation=] len += MAX (0, snprintf (s + len, ARRAY_LENGTH (s) - len, "%g", variation- value)); ~^ /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanag er/harfbuzz/hb-common.cc:1066:27: note: assuming directive output of 2147488582 bytes cc1plus: all warnings being treated as errors (build is a product - build) Setting –disable-warnings-as-errors works as a workaround , but of course this is not really what we want to do . Fixing in theharfbuzz sources in OpenJDk might be also not so nice because it would clash with imports of new versions of harfbuzz . Do you think we could disable the specific warning for the library compilation ? Any other great suggestions 😉 ? Thanks, Matthias
RE: gcc 7.3.1 build - warnings as errors in harfbuzz
Hi Erik, so I think I could disable the warning here : Awt2dLibraries.gmk - $(eval $(call SetupJdkLibrary, BUILD_LIBFONTMANAGER, \ NAME := fontmanager, \ WARNINGS_AS_ERRORS_xlc := false, \ DISABLED_WARNINGS_gcc := format-truncation sign-compare int-to-pointer-cast \ type-limits missing-field-initializers implicit-fallthrough \ strict-aliasing undef unused-function, \ DISABLED_WARNINGS_CXX_gcc := format-truncation reorder delete-non-virtual-dtor strict-overflow \ maybe-uninitialized, \ ... (add format-truncationfor gcc in the warning-disabling section). However this would disable it also for other versions of gcc where the issue never showed up . Do you think this is fine (and safe for older gcc) ? Best regards, Matthias > -Original Message- > From: Erik Joelsson > Sent: Mittwoch, 10. Oktober 2018 17:33 > To: Baesken, Matthias ; 'build- > d...@openjdk.java.net' > Subject: Re: gcc 7.3.1 build - warnings as errors in harfbuzz > > In this case, disabling the warning seems like the right thing to do. > > /Erik > > > On 2018-10-10 06:14, Baesken, Matthias wrote: > > Hello , when compiling jdk/jdk with gcc 7.3.1on linux x86_64 (or > > also > on linux ppc64) I run into this build error : > > > > > > > /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanag > er/harfbuzz/hb-common.cc: In function 'void > hb_variation_to_string(hb_variation_t*, char*, unsigned int)': > > > /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanag > er/harfbuzz/hb-common.cc:1066:27: error: '%g' directive output between 1 > and 18446744073709551615 bytes may cause result to exceed 'INT_MAX' [- > Werror=format-truncation=] > > len += MAX (0, snprintf (s + len, ARRAY_LENGTH (s) - len, "%g", > > variation- > >value)); > > > ~^ > > > > /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanag > er/harfbuzz/hb-common.cc:1066:27: note: assuming directive output of > 2147488582 bytes > > cc1plus: all warnings being treated as errors > > > > (build is a product - build) > > > > Setting –disable-warnings-as-errors works as a workaround , but of > course this is not really what we want to do . > > > > Fixing in theharfbuzz sources in OpenJDk might be also not so nice > because it would clash with imports of new versions of harfbuzz . > > Do you think we could disable the specific warning for the library > compilation ? > > > > Any other great suggestions 😉 ? > > > > > > Thanks, Matthias
Re: gcc 7.3.1 build - warnings as errors in harfbuzz
In this case, disabling the warning seems like the right thing to do. /Erik On 2018-10-10 06:14, Baesken, Matthias wrote: Hello , when compiling jdk/jdk with gcc 7.3.1on linux x86_64 (or also on linux ppc64) I run into this build error : /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanager/harfbuzz/hb-common.cc: In function 'void hb_variation_to_string(hb_variation_t*, char*, unsigned int)': /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanager/harfbuzz/hb-common.cc:1066:27: error: '%g' directive output between 1 and 18446744073709551615 bytes may cause result to exceed 'INT_MAX' [-Werror=format-truncation=] len += MAX (0, snprintf (s + len, ARRAY_LENGTH (s) - len, "%g", variation->value)); ~^ /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanager/harfbuzz/hb-common.cc:1066:27: note: assuming directive output of 2147488582 bytes cc1plus: all warnings being treated as errors (build is a product - build) Setting –disable-warnings-as-errors works as a workaround , but of course this is not really what we want to do . Fixing in theharfbuzz sources in OpenJDk might be also not so nice because it would clash with imports of new versions of harfbuzz . Do you think we could disable the specific warning for the library compilation ? Any other great suggestions 😉 ? Thanks, Matthias
gcc 7.3.1 build - warnings as errors in harfbuzz
Hello , when compiling jdk/jdk with gcc 7.3.1on linux x86_64 (or also on linux ppc64) I run into this build error : /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanager/harfbuzz/hb-common.cc: In function 'void hb_variation_to_string(hb_variation_t*, char*, unsigned int)': /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanager/harfbuzz/hb-common.cc:1066:27: error: '%g' directive output between 1 and 18446744073709551615 bytes may cause result to exceed 'INT_MAX' [-Werror=format-truncation=] len += MAX (0, snprintf (s + len, ARRAY_LENGTH (s) - len, "%g", variation->value)); ~^ /open_jdk/jdk_just_clone/jdk/src/java.desktop/share/native/libfontmanager/harfbuzz/hb-common.cc:1066:27: note: assuming directive output of 2147488582 bytes cc1plus: all warnings being treated as errors (build is a product - build) Setting –disable-warnings-as-errors works as a workaround , but of course this is not really what we want to do . Fixing in theharfbuzz sources in OpenJDk might be also not so nice because it would clash with imports of new versions of harfbuzz . Do you think we could disable the specific warning for the library compilation ? Any other great suggestions 😉 ? Thanks, Matthias
