I have to agree that that isn't the most efficient method for keeping track of the "errorlevel_changed". I will be resubmitting the patch with a better implementation, using the process stack as you suggested.
However, I would like to say that it really wouldn't be possible to only use the success/failure value from the functions that don't change the "errorlevel". This is because there is a subtle difference that sadly Windows thought was a great idea (I think it should work like you say). Windows checks if the "errorlevel" has changed in *any* of the commands inside of a call and if it has then it checks the errorlevel, if it hasn't changed it assumes a success. This means that even if the last command "succeeds" without changing the errorlevel it doesn't matter (e.g. "echo test") if any of the previous commands inside of a call updated the errorlevel. Working examples can be seen on line 291-312 of programs/cmd/tests/test-builtins.cmd or a small snippet of them below:
-snip
:cfail echo %1 call :setError 1 goto :eof
-snip-
call :cfail c1||call :cfail c2
-snip-
:setError exit /B %1
-snip-
The code above will return (and would do so even if one inserted a "echo foobar" after the "call :setError 1"): c1 c2
On Tue, Sep 19, 2017 at 5:27 AM, Alexandre Julliard julliard@winehq.org wrote:
Alexander Coffin alexcoffin1999@gmail.com writes:
@@ -1478,7 +1557,9 @@ void WCMD_execute (const WCHAR *command, const
WCHAR *redirects,
switch (i) { case WCMD_CALL:
push_errorlevel_changed(); WCMD_call (p);pop_errorlevel_changed();This seems to be the only place that does this, so I don't see why you need to manage your own stack instead of saving/restoring things on the process stack.
Also returning a success/failure value from the various functions that implement commands would probably be cleaner than adding a global 'changed' flag.
-- Alexandre Julliard julliard@winehq.org