[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20170615000005.GJ1627@brightrain.aerifal.cx>
Date: Wed, 14 Jun 2017 20:00:05 -0400
From: Rich Felker <dalias@...c.org>
To: musl@...ts.openwall.com
Subject: Re: [PATCH 2/3] confstr: don't give empty strings for unset
values
On Fri, Jun 09, 2017 at 12:26:17AM -0500, A. Wilcox wrote:
> From: "A. Wilcox" <AWilcox@...cox-Tech.com>
>
> If confstr(3) returns an empty string, that means the implementation
> supports the configuration and simply needs no setting. Some values
> should actually be unset on musl. This change ensures that
> configurations not supported by musl do not have values.
> ---
> src/conf/confstr.c | 13 +++++++++++++
> 1 file changed, 13 insertions(+)
>
> diff --git a/src/conf/confstr.c b/src/conf/confstr.c
> index 02cb1aa..c899abd 100644
> --- a/src/conf/confstr.c
> +++ b/src/conf/confstr.c
> @@ -11,6 +11,19 @@ size_t confstr(int name, char *buf, size_t len)
> errno = EINVAL;
> return 0;
> }
> +
> + if (name == _CS_POSIX_V7_WIDTH_RESTRICTED_ENVS ||
> + // Since sysconf _SC_V7_ILP32_OFF32 = -1
> + name == _CS_POSIX_V7_ILP32_OFF32_CFLAGS ||
> + name == _CS_POSIX_V7_ILP32_OFF32_LDFLAGS ||
> + name == _CS_POSIX_V7_ILP32_OFF32_LIBS ||
> + // Since sysconf _SC_V7_LPBIG_OFFBIG = -1
> + name == _CS_POSIX_V7_LPBIG_OFFBIG_CFLAGS ||
> + name == _CS_POSIX_V7_LPBIG_OFFBIG_LDFLAGS ||
> + name == _CS_POSIX_V7_LPBIG_OFFBIG_LIBS) {
> + return 0;
> + }
I'm not sure musl is actually correct in reporting these right now.
POSIX describes LPBIG_OFFBIG as long, pointer, and off_t being "at
least 64 bits", which is the case on 64-bit archs. Of course
ILP32_OFF32 is always false. But I don't think POSIX imposes any
requirement that we return an error here anyway. See unistd.h:
_CS_POSIX_V7_ILP32_OFF32_CFLAGS
If sysconf(_SC_V7_ILP32_OFF32) returns -1, the meaning of this
value is unspecified.
^^^^^^^^^^^
Also I'm not sure what makes sense to return for
_CS_POSIX_V7_WIDTH_RESTRICTED_ENVS. On most archs, the only
compilation environment does satisfy the conditions, but it doesn't
have a name that can be passed to getconf -v. On some (x32, maybe
n32?) archs, strictly speaking we're nonconforming by not having such
an environment (I think at least suseconds_t is wrong, maybe blksize_t
too). The requirements for interaction with POSIX.2 (xcu c99 and
getconf commands) make this whole area a mess to implement since we
don't control that part of the implementation.
BTW your other two patches look fine, committing them.
Rich
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
