maintainer-feedback requested: [Bug 202660] Fix lang/ruby21 installed header to fix signed shift warnings for other ports (e.g. devel/rubygem-thrift)
bugzilla-noreply at freebsd.org
bugzilla-noreply at freebsd.org
Tue Aug 25 18:47:19 UTC 2015
Dimitry Andric <dim at FreeBSD.org> has reassigned Bugzilla Automation
<bugzilla at FreeBSD.org>'s request for maintainer-feedback to ruby at FreeBSD.org:
Bug 202660: Fix lang/ruby21 installed header to fix signed shift warnings for
other ports (e.g. devel/rubygem-thrift)
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=202660
--- Description ---
Created attachment 160356
--> https://bugs.freebsd.org/bugzilla/attachment.cgi?id=160356&action=edit
Fix signed left-shifting in lang/ruby21's public ruby.h
During the exp-run in bug 201377, it was found that devel/rubygem-thrift gives
errors with a recent clang 3.7.0 snapshot:
http://package18.nyi.freebsd.org/data/headi386PR201377-default/2015-08-20_15h42
m20s/logs/errors/rubygem-thrift-0.9.1,1.log
This is because it includes ruby.h from the lang/ruby21 port, which uses this
macro:
#define INT2FIX(i) ((VALUE)(((SIGNED_VALUE)(i))<<1 | FIXNUM_FLAG))
Unfortunately, left-shifting negative 'i' values is undefined behavior, so
clang warns about this. I changed the macro to:
#define INT2FIX(i) ((VALUE)(((VALUE)(i))<<1 | FIXNUM_FLAG))
which avoids the undefined behavior by casting to VALUE (which is unsigned)
first. I ran all ruby21 tests before and after this fix, and I got "1
failures, 4 errors" in both cases, so no regressions.
However, since this is a public ruby header, I can imagine that this is a
change that makes some people nervous. So if ruby maintainers prefer to fix
this in the devel/rubygem-thrift port instead, for example by squelching the
warning, please let me know.
More information about the freebsd-ruby
mailing list