Bug 96245 – [Qt] Fix generation of forward headers for generated files in WebCore on Windows

• Component: New Bugs
• Status: RESOLVED FIXED

[Simon Hausmann]

[Qt] Fix generation of forward headers for generated files in WebCore on Windows

[Simon Hausmann]

Created attachment 163063 [details]
Patch

[Tor Arne Vestbø]

Comment on attachment 163063 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=163063&action=review

> Source/WebKit2/DerivedSources.pri:148
> +    win32: eval($${header_target}.commands = ($${QMAKE_MKDIR} $$toSystemPath($$dest_dir) 2>nul || echo>nul))

It looks like we use $${QMAKE_CHK_DIR_EXISTS} || $${QMAKE_MKDIR} in the rest of Qt's makefiles, perhaps stick with that?

[Simon Hausmann]

Comment on attachment 163063 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=163063&action=review

>> Source/WebKit2/DerivedSources.pri:148
>> +    win32: eval($${header_target}.commands = ($${QMAKE_MKDIR} $$toSystemPath($$dest_dir) 2>nul || echo>nul))
> 
> It looks like we use $${QMAKE_CHK_DIR_EXISTS} || $${QMAKE_MKDIR} in the rest of Qt's makefiles, perhaps stick with that?

The problem with that approach is that it doesn't work well with parallel rules that try to create the same directory ;(

IOW there's a race condition between the check if the directory exists and making it. If you have two jobs trying to do the same thing, then
both jobs determine simultaneously that the directory doesn't exist, both try to create the directory at the same time and only once succeeds. The
second one will see mkdir aborting because the directory already exists.

I tried removing the mkdir from this rule altogether and instead depend on the other generator that creates the directory also, but unfortunately when
depending on such a symbol target nmake will always call it and thus always re-generate the file.

[Tor Arne Vestbø]

Comment on attachment 163063 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=163063&action=review

>>> Source/WebKit2/DerivedSources.pri:148
>>> +    win32: eval($${header_target}.commands = ($${QMAKE_MKDIR} $$toSystemPath($$dest_dir) 2>nul || echo>nul))
>> 
>> It looks like we use $${QMAKE_CHK_DIR_EXISTS} || $${QMAKE_MKDIR} in the rest of Qt's makefiles, perhaps stick with that?
> 
> The problem with that approach is that it doesn't work well with parallel rules that try to create the same directory ;(
> 
> IOW there's a race condition between the check if the directory exists and making it. If you have two jobs trying to do the same thing, then
> both jobs determine simultaneously that the directory doesn't exist, both try to create the directory at the same time and only once succeeds. The
> second one will see mkdir aborting because the directory already exists.
> 
> I tried removing the mkdir from this rule altogether and instead depend on the other generator that creates the directory also, but unfortunately when
> depending on such a symbol target nmake will always call it and thus always re-generate the file.

Ouch! You're right, that will not be pretty. Wonder why Qt gets away with that though.

[WebKit Review Bot]

Comment on attachment 163063 [details]
Patch

Clearing flags on attachment: 163063

Committed r128045: <http://trac.webkit.org/changeset/128045>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.