Fix Windows dependency tracking related to JSInternals.h/cpp

• Component: Tools / Tests
• Status: RESOLVED FIXED

[Csaba Osztrogonác]

Windows EWS were stucked for 2.5 days after http://trac.webkit.org/changeset/174315 with this error:

 1>c1xx : fatal error C1083: Cannot open source file: 'C:\cygwin\home\buildbot\WebKit\WebKitBuild\Release\obj32\WebCore\DerivedSources\JSInternals.cpp': No such file or directory
     1>..\testing\js\WebCoreTestSupport.cpp(33): fatal error C1083: Cannot open include file: 'JSInternals.h': No such file or directory
     1>Done Building Project "C:\cygwin\home\buildbot\WebKit\Source\WebCore\WebCore.vcxproj\WebCoreTestSupport.vcxproj" (Build target(s)) -- FAILED.


Buildfix attempt / workaround landed in r174344. (Similar patch landed in r144570 to fix a similar strange Win failure.)

[Csaba Osztrogonác]

It seems r174344 didn't help. It would be great if anybody 
from Apple can fix this bug to unbreak the Windows EWS bots.

[Csaba Osztrogonác]

One more attempt from Roger - http://trac.webkit.org/changeset/174378 ,
but the EWS is still broken.

[Csaba Osztrogonác]

Fixed by https://trac.webkit.org/changeset/174541.

Just out of curiosity, could you tell us how is it possible that 
the problem happened on the EWS bots only, but not on the buildbots?

What was the exact problem why we needed to move the enum declaration
to the top of the file? If it is design problem in the generator or
if it is a Windows issue, we should add a new rule to the idl style
checker to force folks to put the enum declaration to the proper place.

[Tim Horton]

Ha! Yeah, I'd really like to understand this better.

[Csaba Osztrogonác]

(In reply to comment #2)
> One more attempt from Roger - http://trac.webkit.org/changeset/174378 ,
> but the EWS is still broken.

Roger, could you explain us, why this unreviewed patch fixed the issue?
What was the problem, and why was this the proper fix?

[Csaba Osztrogonác]

(In reply to comment #5)
> (In reply to comment #2)
> > One more attempt from Roger - http://trac.webkit.org/changeset/174378 ,
> > but the EWS is still broken.
> 
> Roger, could you explain us, why this unreviewed patch fixed the issue?
> What was the problem, and why was this the proper fix?

I meant r174541.

[Csaba Osztrogonác]

ping

[Csaba Osztrogonác]

Roger, please don't ignore our reqeust to explain why your unreviewed
and uncommented patch - http://trac.webkit.org/changeset/174541 fixed
the Windows issue. Please answer ASAP. Thanks.

[Roger Fong]

(In reply to comment #8)
> Roger, please don't ignore our reqeust to explain why your unreviewed
> and uncommented patch - http://trac.webkit.org/changeset/174541 fixed
> the Windows issue. Please answer ASAP. Thanks.

Apologies, I've neglected my bugzilla inbox recently. Added a few too many people to my watch list and it's been getting ultra flooded.

Unfortunately I don't have a particularly good answer for you. 
The only hunch that I have is that it might have to do with needing to upgrade the vs2013 express service packs on those machines but I found my actual solution via a sort of last resort guess and check. The next chance I get I'll try to upgrade those bots (which I have to manually right now), locally revert my fix, and see if the problem still occurs.

[Csaba Osztrogonác]

(In reply to comment #9)
> (In reply to comment #8)
> > Roger, please don't ignore our reqeust to explain why your unreviewed
> > and uncommented patch - http://trac.webkit.org/changeset/174541 fixed
> > the Windows issue. Please answer ASAP. Thanks.
> 
> Apologies, I've neglected my bugzilla inbox recently. Added a few too many
> people to my watch list and it's been getting ultra flooded.

No problem. You can setup a more sophisticated mail filter
for bugzilla mails to separate important / non-important mails.
There are many reason why do you get a bugzilla mail, you can 
find them at the end of the mail: "You are receiving this mail because:"

Here is the list of these reasons:
http://trac.webkit.org/browser/trunk/Websites/bugs.webkit.org/template/en/default/global/reason-descs.none.tmpl?order=name#L20

> Unfortunately I don't have a particularly good answer for you. 
> The only hunch that I have is that it might have to do with needing to
> upgrade the vs2013 express service packs on those machines but I found my
> actual solution via a sort of last resort guess and check. The next chance I
> get I'll try to upgrade those bots (which I have to manually right now),
> locally revert my fix, and see if the problem still occurs.

It's strange. Maybe not r174541 fixed the issue, but some tweak on the servers.
Otherwise it would be great to have exactly the same configuration on the EWS
and the buildbots to avoid similar issues in the future.