Re: [PATCH v4] win32: fix thread usage for win32

[Johannes Sixt <j6t@kdbg.org>]

Am 23.01.23 um 18:43 schrieb Jeff Hostetler:
> 
> 
> On 1/23/23 11:48 AM, Rose via GitGitGadget wrote:
>> From: Seija Kijin <doremylover123@gmail.com>
>>
>> Use _beginthreadex instead of CreateThread
>> since we use the Windows CRT,
>> as Microsoft recommends _beginthreadex
>> over CreateThread for these situations.
>>
>> Finally, check for NULL handles, not "INVALID_HANDLE,"
>> as _beginthreadex guarantees a valid handle in most cases
>>
>> Signed-off-by: Seija Kijin <doremylover123@gmail.com>
>> ---
>>      win32: fix thread usage for win32
>>           Use pthread_exit instead of async_exit.
>>           This means we do not have to deal with Windows's implementation
>>      requiring an unsigned exit coded despite the POSIX exit code
>> requiring a
>>      signed exit code.
>>           Use _beginthreadex instead of CreateThread since we use the
>> Windows CRT.
>>           Finally, check for NULL handles, not "INVALID_HANDLE," as
>> _beginthreadex
>>      guarantees a valid handle in most cases
>>           Signed-off-by: Seija Kijin doremylover123@gmail.com
>>
>> Published-As:
>> https://github.com/gitgitgadget/git/releases/tag/pr-git-1440%2FAtariDreams%2FCreateThread-v4
>> Fetch-It-Via: git fetch https://github.com/gitgitgadget/git
>> pr-git-1440/AtariDreams/CreateThread-v4
>> Pull-Request: https://github.com/git/git/pull/1440
>>
>> Range-diff vs v3:
>>
>>   1:  68baafba2bd ! 1:  2e2d5ce7745 win32: fix thread usage for win32
>>       @@ Commit message
>>                   Signed-off-by: Seija Kijin <doremylover123@gmail.com>
>>              - ## compat/mingw.c ##
>>       -@@ compat/mingw.c: static int start_timer_thread(void)
>>       -     timer_event = CreateEvent(NULL, FALSE, FALSE, NULL);
>>       -     if (timer_event) {
>>       -         timer_thread = (HANDLE) _beginthreadex(NULL, 0,
>> ticktack, NULL, 0, NULL);
>>       --        if (!timer_thread )
>>       -+        if (!timer_thread)
>>       -             return errno = ENOMEM,
>>       -                 error("cannot start timer thread");
>>       -     } else
>>       -
>>         ## compat/winansi.c ##
>>        @@ compat/winansi.c: enum {
>>             TEXT = 0, ESCAPE = 033, BRACKET = '['
>>
>>
>>   compat/winansi.c | 8 ++++----
>>   1 file changed, 4 insertions(+), 4 deletions(-)
>>
>> diff --git a/compat/winansi.c b/compat/winansi.c
>> index 3abe8dd5a27..be65b27bd75 100644
>> --- a/compat/winansi.c
>> +++ b/compat/winansi.c
>> @@ -340,7 +340,7 @@ enum {
>>       TEXT = 0, ESCAPE = 033, BRACKET = '['
>>   };
>>   -static DWORD WINAPI console_thread(LPVOID unused)
>> +static unsigned int WINAPI console_thread(LPVOID unused)
>>   {
>>       unsigned char buffer[BUFFER_SIZE];
>>       DWORD bytes;
>> @@ -643,9 +643,9 @@ void winansi_init(void)
>>           die_lasterr("CreateFile for named pipe failed");
>>         /* start console spool thread on the pipe's read end */
>> -    hthread = CreateThread(NULL, 0, console_thread, NULL, 0, NULL);
>> -    if (hthread == INVALID_HANDLE_VALUE)
>> -        die_lasterr("CreateThread(console_thread) failed");
>> +    hthread = (HANDLE)_beginthreadex(NULL, 0, console_thread, NULL,
>> 0, NULL);
>> +    if (!hthread)
>> +        die_lasterr("_beginthreadex(console_thread) failed");
>>         /* schedule cleanup routine */
>>       if (atexit(winansi_exit))
>>
>> base-commit: 56c8fb1e95377900ec9d53c07886022af0a5d3c2
> 
> This change may or may not be harmless, but it scares me
> because it is possibly a very subtle change and is being
> made for an unknown reason -- is there a problem being
> fixed here?  Or is this just churn for the sake of churn
> to avoid an awkward cast of the return code?
> 
> What does _beginthreadex() specifically do that we need
> it to do for us?
> 
> _beginthreadex() does some CRT init and then calls CreateThread(),
> so what are we missing by calling CreateThread() directly?

I also question the value of this change. As long as the thread does not
call into any CRT functions, we do not need the services of
_beginthreadex(). AFAICS, it only uses WinAPI functions and some
uncritical C functions like memmove and memset. Am I missing something?

> 
> The code in question is 11+ years old and it hasn't been a
> problem (right?), so I have to wonder what value do we get
> from this change.
> 
> The containing function here is setting up a special console
> thread and named pipe to access the console, so I doubt that
> any of the tests in the test suite actually would actually
> exercise this change (since the tests aren't interactive).
> 
> The low-level Windows startup code is very tricky and sensitive
> (and we need to test with both GCC's CRT and MSVC's CRT).
> As I said earlier, the change may or may not be harmless, but
> I question the need for it.
> 
> Jeff
> 
>