From: Martin Sebor via Libc-alpha <libc-alpha@sourceware.org>
To: Joseph Myers <joseph@codesourcery.com>
Cc: Florian Weimer <fweimer@redhat.com>,
GNU C Library <libc-alpha@sourceware.org>
Subject: Ping: [PATCH] add support for -Wmismatched-dealloc
Date: Mon, 4 Jan 2021 08:56:01 -0700 [thread overview]
Message-ID: <572110cb-70f7-fd85-bdea-b95eec05e9b8@gmail.com> (raw)
In-Reply-To: <655918b2-16c6-74b1-6a49-505a7607007f@gmail.com>
Florian/Joseph and/or others: is the latest patch okay to commit?
https://sourceware.org/pipermail/libc-alpha/2020-December/121121.html
On 12/27/20 4:13 PM, Martin Sebor wrote:
> More testing made me realize that further changes are needed:
> 1) correct the return value of the __fclose() alias to int,
> 2) declare and use the same alias for fclose in both <stdio.h>
> and <wchar.h>.
>
> In addition, I noticed a few more opportunities to use the new
> attribute:
> * in include/programs/xmalloc.h,
> * in malloc/malloc.h,
> * and in wcsdup in <wchar.h>.
>
> I also simplified the new macro definitions a bit, and added
> a new test to verify that the warning doesn't cause false
> positives for open_wmemstream.
>
> Attached is a patch with these updates.
>
> On 12/15/20 9:52 AM, Martin Sebor wrote:
>> On 12/14/20 6:01 PM, Joseph Myers wrote:
>>> On Mon, 14 Dec 2020, Martin Sebor via Libc-alpha wrote:
>>>
>>>>> I spent some time working around this but in the end it turned out
>>>>> to be too convoluted so I decided to make the attribute a little
>>>>> smarter. Instead of associating all allocation functions with all
>>>>> deallocation functions (such as fdopen, fopen, fopen64, etc. with
>>>>> fclose, freopen, and freopen64) I changed it so that an allocator
>>>>> only needs to be associated with a single deallocator (a reallocator
>>>>> also needs to be associated with itself). That makes things quite
>>>>> a bit simpler.
>>> [...]
>>>> The GCC patches have now been committed and the dependency resolved.
>>>
>>> I've looked at the attribute documentation now in GCC, but I'm afraid
>>> I'm
>>> unable to understand from that documentation why the proposed glibc
>>> patch
>>> constitutes a valid way of specifying that, for example, it's valid
>>> to use
>>> freopen as a deallocator for FILE pointers opened by functions whose
>>> attribute only mentions fclose. Unless there's something I'm missing in
>>> the documentation or a separate documentation patch that's not yet
>>> committed, I think more work is needed on the GCC documentation to make
>>> clear the semantics the glibc patch is asserting for valid
>>> combinations of
>>> allocators and deallocators, so that those semantics can be reviewed for
>>> correctness.
>>
>> I flip-flopped with freopen. Initially I wanted to mark it up as
>> both an allocator and a deallocator, analogously to realloc (which
>> is implicitly both) or reallocarray (which is annotated as both in
>> the latest Glibc patch). Both the initial Glibc and GCC patches
>> (the manual for the latter) reflected this and had freopen annotated
>> that way.
>>
>> But because freopen doesn't actually deallocate or allocate a stream
>> the markup wouldn't be correct. It would cause false positives with
>> -Wmismatched-dealloc as well with other warnings like the future
>> -Wuse-after-free (or with -Wanalyzer-use-after-free when the GCC
>> analyzer adds support for the attribute that David Malcolm is
>> working on for GCC 11). I've added a test case to the test suite:
>>
>> void f (FILE *f1)
>> {
>> FILE *f2 = freopen ("", "", f1);
>> fclose (f1); // must not warn
>> }
>>
>> To answer your question, without the attribute freopen is seen by
>> GCC as an ordinary function that happens to take a FILE* and return
>> another FILE*. It neither allocates it nor deallocates it. For
>> GCC 12, I'd like us to consider adding attribute returns_arg(position)
>> to improve the analysis here. The GCC manual also doesn't mention
>> freopen anymore but I'd be happy to change the example there to
>> show an API that does include a reallocator (e.g., reallocarray).
>>
>> Having said all this, after double-checking the latest Glibc patch
>> I see it still has the attribute on freopen by mistake (as well as
>> the ordinary attribute malloc, which would make it even worse).
>> I've removed both in the attached revision. Sorry if this confused
>> you -- freopen obviously confused me.
>>
>> Martin
>
next prev parent reply other threads:[~2021-01-04 15:56 UTC|newest]
Thread overview: 27+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-12-08 22:52 [PATCH] add support for -Wmismatched-dealloc Martin Sebor via Libc-alpha
2020-12-09 0:07 ` Joseph Myers
2020-12-12 2:25 ` Martin Sebor via Libc-alpha
2020-12-14 21:39 ` Martin Sebor via Libc-alpha
2020-12-14 22:16 ` Florian Weimer via Libc-alpha
2020-12-15 1:01 ` Joseph Myers
2020-12-15 16:52 ` Martin Sebor via Libc-alpha
2020-12-27 23:13 ` Martin Sebor via Libc-alpha
2021-01-04 15:56 ` Martin Sebor via Libc-alpha [this message]
2021-01-04 16:07 ` Florian Weimer via Libc-alpha
2021-01-04 16:18 ` Martin Sebor via Libc-alpha
2021-01-04 16:57 ` Florian Weimer via Libc-alpha
2021-01-04 23:18 ` Martin Sebor via Libc-alpha
2021-01-10 20:42 ` Ping: " Martin Sebor via Libc-alpha
2021-01-11 9:13 ` Florian Weimer via Libc-alpha
2021-01-12 0:00 ` Martin Sebor via Libc-alpha
2021-01-12 0:01 ` Martin Sebor via Libc-alpha
2021-01-12 8:59 ` Florian Weimer via Libc-alpha
2021-01-19 1:08 ` Martin Sebor via Libc-alpha
2021-01-19 16:54 ` David Malcolm via Libc-alpha
2021-01-22 21:26 ` DJ Delorie via Libc-alpha
2021-01-25 10:56 ` Florian Weimer via Libc-alpha
2021-01-25 11:31 ` Florian Weimer via Libc-alpha
2021-04-23 0:00 ` Martin Sebor via Libc-alpha
2021-05-06 23:54 ` Martin Sebor via Libc-alpha
2021-05-13 21:49 ` Martin Sebor via Libc-alpha
2021-05-16 21:25 ` Martin Sebor via Libc-alpha
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
List information: https://www.gnu.org/software/libc/involved.html
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=572110cb-70f7-fd85-bdea-b95eec05e9b8@gmail.com \
--to=libc-alpha@sourceware.org \
--cc=fweimer@redhat.com \
--cc=joseph@codesourcery.com \
--cc=msebor@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).