560723 - Sort out why the _Throw() hack doesn't work on VC10, and find an alternative

Status: RESOLVED FIXED

(Core :: XPCOM, defect)

Comment 1

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Attachment: Interpose mozilla variants of functions that throw exceptions from the VC10 STL

VC10 departs from the VC9/8/7 exception-throwing strategy pretty significantly.  There appear to be two ways "exceptions" are "thrown": (i) by a _RAISE() macro that calls _invoke_watson() in _HAS_EXCEPTIONS=0 code; (ii) by _X[exception]() functions exported by the CRT, e.g. _Xout_of_range() (like gcc).  (i) isn't so bad, but (ii) results in C++ exceptions being thrown from the CRT (since it's _HAS_EXCEPTIONS=1), and that's bad.  This patch again grossly interposes our definitions of these, and part of me died as usual.

That said, if we want to switch releasing VC builds, we might want this patch eventually for another reason: whether or not |if _HAS_EXCEPTIONS|, xul.dll will retain a link-time dependency on the _X[exception] functions.  They're not exported by the <=VC9 CRT, which means that eventually the successor of bug 557060 would be filed.  Small benefit, yes, but hopefully it'll wash out some of the mouth-vom.

Comment 2

[(no longer active)]

Comment on attachment 440710 [details] [diff] [review]
Interpose mozilla variants of functions that throw exceptions from the VC10 STL

I'm sad that we have to do this, but I think this is the best we can currently do.  r=me.

Comment 3

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Attachment: hg export'd version of above

Comment 4

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

http://hg.mozilla.org/mozilla-central/rev/6e7fe366cab7

Comment 5

[Robert Kaiser]

This checkin caused Thunderbird/SeaMonkey breakage, from a fast glance I'd say LDAP needs some change to build with this, but I'm not digging deeper as Standard8 has already been looking into it and says he probably has a fix but still needs to test it.