Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 14 Feb 2016 05:47:19 +1100 (EST)
From:      Bruce Evans <brde@optusnet.com.au>
To:        Konstantin Belousov <kostikbel@gmail.com>
Cc:        Bruce Evans <brde@optusnet.com.au>, src-committers@freebsd.org,  svn-src-all@freebsd.org, svn-src-head@freebsd.org
Subject:   Re: svn commit: r295561 - in head: include sys/mips/include sys/powerpc/include sys/sparc64/include sys/sys sys/x86/include
Message-ID:  <20160214050522.J1687@besplex.bde.org>
In-Reply-To: <20160213172321.GE91220@kib.kiev.ua>
References:  <201602120738.u1C7cKpq093956@repo.freebsd.org> <20160212232717.P894@besplex.bde.org> <20160212143630.GS91220@kib.kiev.ua> <20160213021939.S1340@besplex.bde.org> <20160212173641.GV91220@kib.kiev.ua> <20160213070506.N2501@besplex.bde.org> <20160213101623.GZ91220@kib.kiev.ua> <20160214035852.D918@besplex.bde.org> <20160213172321.GE91220@kib.kiev.ua>

next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, 13 Feb 2016, Konstantin Belousov wrote:

> On Sun, Feb 14, 2016 at 04:02:05AM +1100, Bruce Evans wrote:
>> The old visibility bugs with mc_* should be fixed someday.  I said that
>> uc_* could be used.  _mc_* should have been used originally.  Maybe we
>> can just change to either of these, since nothing except libc should
>> have used the internals of mcontext_t.
>
> mcontext_t is needed and used by many language runtimes which handle
> exceptions for their functionality. It could be gc barriers, hacks with
> tagging, and many more. API cannot be broken there by renaming the
> structure members.

But it is not part of the API.  The API is specified to be opaque.
getcontext(3) doesn't even document the visible parts of it.

> The only way to hide mc_* is to rename them to __mc_* and provide compat
> redefinitions when <machine/ucontext.h> is included directly. This is
> extremely cumbersome and I do not see a need in such fix for perceived
> use case of pure ANSI C code which, to be broken, must #define mc_XXX
> and then include <signal.h>.

How likely is third party code to use undocumented internals of opqaue
types when FreeBSD code mostly doesn't used them.

I checked all files in /usr/src outside of /sys/ that contain "ucontext.h".
There are 71 such files with 204 lines matching "mc_".  The actual users
of mcontext_t's internals are:

contrib/compiler-rt/lib/sanitizer_common/sanitizer_linux.cc: 8 instances
contrib/compiler-rt/lib/sanitizer_common/sanitizer_freebsd.h: 30 instances
lib/libc: 102 instances
lib/libthread_db: 60 instances
tools/KSE/ksetest/kse_threads_test.c: 2 instances
tools/KSE/rr/rr.c: 2 instances

So there are 4 instances outside of libraries (counting compiler_rt as
a libraries) and these 4 are to support KSE which went away about 11
years ago IIRC.  tools/KSE doesn't compile now of course.  The first
error is that its asm source file is i386-only, so fails on i386.
The next error is that its primary (?) API file <sys/kse.h> no longer
exists.

I hoped to find no instances outside of libc.  Bits in contrib are are
problem.

Bruce



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20160214050522.J1687>