Windows binary saga

[Dirk Hohndel]

On Wed, Sep 23, 2015 at 12:47:41PM +0300, Lubomir I. Ivanov wrote:
> On 23 September 2015 at 12:28, Lubomir I. Ivanov <neolit123 at gmail.com> wrote:
> > On 23 September 2015 at 10:54, Lubomir I. Ivanov <neolit123 at gmail.com> wrote:
> >> On 23 September 2015 at 06:52, Dirk Hohndel <dirk at hohndel.org> wrote:
> >>>
> >>> Anyway, I'll stop here for today. Feel free to play with it - it's in
> >>> downloads/daily as subsurface-4.4.97-26-ge66f0895c68c.exe
> >>>
> >>
> >> ok, i will give it a spin later today.
> >>
> >
> > Windows 7 SP1, 64bit.
> >
> > modules tested:
> > - location edit - quick test OK
> > - printing  - importing, exporting, previewing OK
> > - cloud storage: crashes
> >
> > the crash is in libgit2.dll at the _git_diff__oid_for_entry() branch.
> > i didn't backtrace that to a public branch, because it refuses to
> > crash under a debugger.
> > the first thing i tried is to replace libgit2.dll and libssh2.dll with
> > the ones i build and it works.
> >
> > my libgit2 is at:
> > 6c21211c3842ae
> >
> > this could be a hotfix that i have, or if you are ahead it could be a
> > regression.
> > either way, i consider this to be a libgit2 issue or something that we
> > do wrong which that particular libgit2 is sensitive to.
> >
> > i would try re-building libgit2 either with my hash version or trying
> > to bisect the cause.
> > i can also send you my DLLs as the backup plan.
> >
> 
> to clarify further,
> 
> if i delete the local cache and then if i let my libgit2.dll create
> it, the libgit2.dll bundled in subsurface-4.4.97-26-ge66f0895c68c.exe
> works fine with said cache.
> but if let the bundled DLL create the cache it crashes.

That's very useful information. Let me make sure I'm using a sane libgit2
version for building this... I moved so many things around that I wouldn't
be surprised if I ended up with an older-than-desired version in my
Windows build...

Thanks for tracking this down.

On Windows 10 you apparently no longer have the option to see a stack
trace or any crash information. It appears to simply send that data to
Microsoft without even an opt-out...

Which will be a real pain for things that don't crash under the debugger.

Or maybe there is a magic incantation that I wasn't able to figure out :-)

/D