Bug #13069
closedmkmf: ignore linker warnings on OpenBSD
Description
Installing gems with native extensions fails on my OpenBSD machine since
the linker emits warnings causing stderr to not be empty as expected.
The warnings are security recommendations that does not affect the
linked binary. See examples below from my mkmf.log caused by libruby:
warning: warning: strcpy() is almost always misused, please use strlcpy()
warning: warning: strcat() is almost always misused, please use strlcat()
warning: warning: sprintf() is often misused, please use snprintf()
warning: warning: vsprintf() is often misused, please use vsnprintf()
Attached is patch that checks the output for such harmless warnings. Not
sure if this exception should be allowed on all platforms, but instead
wrapped in a RUBY_PLATFORM =~ /openbsd/
conditional or something
similar.
Files
Updated by jeremyevans0 (Jeremy Evans) almost 8 years ago
Personally, I think it would be better to just check the return code and ignore the content of the log file. Are there environments that ruby wants to support where compiling returns success when it actually fails? The supplied patch won't handle all of the warnings produced by OpenBSD's linker, though it does handle the most common ones.
Updated by nobu (Nobuyoshi Nakada) almost 8 years ago
- Description updated (diff)
- Status changed from Open to Feedback
Seems sprintf
nor vsprintf
are used on OpenBSD.
Those warnings are always emitted regardless use of the functions?
Updated by Anonymous almost 8 years ago
Nobuyoshi Nakada wrote:
Seems
sprintf
norvsprintf
are used on OpenBSD.
Those warnings are always emitted regardless use of the functions?
Here's a minimal example causing the errors to be outputted while
dynamically linking against ruby:
$ env "PATH=${DESTDIR}/usr/local/bin:${PATH}" ruby -v
ruby 2.4.0dev (2016-12-24 trunk 57169) [x86_64-openbsd6.0]
$ cat test.c
#include "ruby.h"
int main(void) { return 0; }
$ cc \
-o test \
-I ${DESTDIR}/usr/local/include/ruby-2.4.0/x86_64-openbsd6.0 \
-I ${DESTDIR}/usr/local/include/ruby-2.4.0 \
test.c \
-L ${DESTDIR}/usr/local/lib \
-lruby \
-lm \
-lpthread
usr/local/lib/libruby.so: warning: warning: strcpy() is almost always misused, please use strlcpy()
usr/local/lib/libruby.so: warning: warning: strcat() is almost always misused, please use strlcat()
Updated by jeremyevans0 (Jeremy Evans) almost 8 years ago
- Status changed from Feedback to Open
Some of the warnings were addressed in r57189, r57190, and r57191. However, I think we should rethink checking for an empty log file, and rely purely on the return code:
--- lib/mkmf.rb.orig Sun Dec 25 09:49:06 2016
+++ lib/mkmf.rb Sun Dec 25 09:49:51 2016
@@ -388,7 +388,7 @@ module MakeMakefile
result = nil
Logging.postpone do |log|
output = IO.popen(libpath_env, command, &:read)
- result = ($?.success? and File.zero?(log.path))
+ result = $?.success?
output
end
result
The only reason I can think of to not rely purely on the return code is if we want to support environments where the command would return 0 in the case where it failed. Are there such environments?
The approach of the current code is similar to defaulting to -Werror
when compiling C, which is almost always a bad idea when distributing C code unless you know every possible environment you want to support (which ruby does not).
Updated by jeremyevans0 (Jeremy Evans) almost 8 years ago
I think this is a better fix. This keeps the :werror
behavior in Logging::postpone
, it just turns off the use of :werror
by default. Users that want werror behavior by default can set $werror = true
.
--- lib/mkmf.rb.orig
+++ lib/mkmf.rb
@@ -88,6 +88,7 @@ module MakeMakefile
$static = nil
$config_h = '$(arch_hdrdir)/ruby/config.h'
$default_static = $static
+ $werror = false unless defined?($werror)
unless defined? $configure_args
$configure_args = {}
@@ -611,7 +612,7 @@ def with_cppflags(flags)
end
def try_cppflags(flags, opts = {})
- try_header(MAIN_DOES_NOTHING, flags, {:werror => true}.update(opts))
+ try_header(MAIN_DOES_NOTHING, flags, {:werror => $werror}.update(opts))
end
def append_cppflags(flags, *opts)
@@ -633,7 +634,7 @@ def with_cflags(flags)
end
def try_cflags(flags, opts = {})
- try_compile(MAIN_DOES_NOTHING, flags, {:werror => true}.update(opts))
+ try_compile(MAIN_DOES_NOTHING, flags, {:werror => $werror}.update(opts))
end
def append_cflags(flags, *opts)
@@ -655,7 +656,7 @@ def with_ldflags(flags)
end
def try_ldflags(flags, opts = {})
- try_link(MAIN_DOES_NOTHING, flags, {:werror => true}.update(opts))
+ try_link(MAIN_DOES_NOTHING, flags, {:werror => $werror}.update(opts))
end
def append_ldflags(flags, *opts)
@@ -1408,7 +1409,7 @@ def convertible_int(type, headers = nil, opts = nil, &b)
u = "unsigned " if signed > 0
prelude << "extern rbcv_typedef_ foo();"
compat = UNIVERSAL_INTS.find {|t|
- try_compile([prelude, "extern #{u}#{t} foo();"].join("\n"), opts, :werror=>true, &b)
+ try_compile([prelude, "extern #{u}#{t} foo();"].join("\n"), opts, :werror=>$werror, &b)
}
end
if compat
This basically reverts the changes made in r50215 and r50216, except that it allows enabling werror globally via $werror
. The commit messages for both commits do not describe why the code was changed to default to {:werror=>true}
. From looking at the history of mkmf, the werror code was originally added in r30107 to work around declaration conflict warnings, so maybe we want to keep :werror=>true
in the final diff block. r35101 states that the exit code may be lost on mingw, so we may want to default $werror
to true on mingw if that is still the case.
I think we need to change the current default of treating warnings as errors. We should handle broken environments that do not use return codes correctly on an exception basis.
Updated by Anonymous almost 8 years ago
Thansk for addressing the warnings. Your proposed patch solves my problem.
Updated by akihikodaki (Akihiko Odaki) almost 7 years ago
- File mkmf.patch mkmf.patch added
- ruby -v set to ruby 2.5.0p0 (2017-12-25 revision 61468) [x86_64-linux]
The issue is affecting my gem, cld3-ruby.
Build failure on OpenBSD · Issue #16 · akihikodaki/cld3-ruby
Those warnings are always emitted regardless use of the functions?
I reviewed the source code of Binutils and GCC and found no means to suppress the warning.
I understand many link warnings result in runtime failure, but it is just insane to make all warnings errors. Warnings are not errors because they may not result in actual failure.
The impact of this issue is severe. It would disable many native extensions on OpenBSD. On the other hand, though those warnings sound ridiculous, OpenBSD should not be blamed because they are just adding warnings, not errors. mkmf should be fixed.
I have two suggestions to address this kind of issue:
- Redirect linker's stderr to mkmf's stderr, and optionally ask the user if he thinks it is fine to continue.
- Execute the output instead of catching stderr to detect runtime failures if not cross compiling.
I attached a patch to implement suggestion 2.
Updated by nobu (Nobuyoshi Nakada) almost 7 years ago
- Status changed from Open to Closed
Applied in changeset trunk|r62007.
mkmf.rb: ignore linker warnings
- lib/mkmf.rb (try_ldflags): ignore linker warnings. they cause
unexpected failures on OpenBSD. [ruby-core:78827] [Bug #13069]
Updated by nobu (Nobuyoshi Nakada) almost 7 years ago
Could you let us know where those functions are used?
It's better to get rid of them, I think.
Updated by hsbt (Hiroshi SHIBATA) about 6 years ago
- Backport changed from 2.1: UNKNOWN, 2.2: UNKNOWN, 2.3: UNKNOWN to 2.3: UNKNOWN, 2.4: UNKNOWN, 2.5: REQUIRED
Updated by nagachika (Tomoyuki Chikanaga) about 6 years ago
- Backport changed from 2.3: UNKNOWN, 2.4: UNKNOWN, 2.5: REQUIRED to 2.3: UNKNOWN, 2.4: UNKNOWN, 2.5: DONE
ruby_2_5 r64979 merged revision(s) 62007,62024.
Updated by usa (Usaku NAKAMURA) about 6 years ago
- Backport changed from 2.3: UNKNOWN, 2.4: UNKNOWN, 2.5: DONE to 2.3: UNKNOWN, 2.4: DONE, 2.5: DONE
ruby_2_4 r65117 merged revision(s) 62007,62024.