2010/12/28 Pierre Joye wrote:

Hi Pierre,

> It is possible to hook (almost) any APIs for an exe too, at least
> using recent VC/crt (no idea about mingw, which relies on vc6 crt
> anyway). I'm not sure if it is worth the initial effort.

No, currently it is not worth. And even less, taking in account that
only one out of four (debug/release shared/static) build combinations
tests would benefit of it.

There are other areas (i.e. sockfilt.c) which require attention before
this hooking would serve any purpose.

>> Debug builds override system's gethostname() using CURL_GETHOSTNAME
>> environment variable, no problem here with this on Windows. So debug
>> builds would should be capable of running the 20+ tests that require
>> system's gethostname() overriding even on Windows.
>
> That should be enough. Or do you see areas where testing using a
> release build is required? in threaded environments maybe, are there
> any tests to test concurency/etc.?

Testing release builds, which are going to be deployed, is always a
good thing. Besides that, debug build tests exercise lib/curls
capabilities at least equally as release builds.

Hmm, now that you mention all this, it remembers me that msvc debug
builds have debug memory tracing disabled intentionally long time ago.
But I can't remember the runtime problem that forced this setting. If
you feel so, you could make a msvc test build with both DEBUGBUILD and
CURLDEBUG (see docs/internals for a bit of info) and try the resulting
curl and libcurl. The problem could be related with existing
makefiles, or who knows what, I really don't remember.

-- 
-=[Yang]=-
-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette:  http://curl.haxx.se/mail/etiquette.html