[PATCH] Set a relative XSLT path for Windows builds

Dirk Hohndel dirk at hohndel.org
Fri Feb 22 20:15:22 PST 2013


"Lubomir I. Ivanov" <neolit123 at gmail.com> writes:

> On 22 February 2013 21:24, Dirk Hohndel <dirk at hohndel.org> wrote:
>> Dirk Hohndel <dirk at hohndel.org> writes:
>>
>>> This is going into the right direction but it still doesn't work as
>>> intended. I've instrumented the code to raise errors that I can see
>>> (stderr goes nowhere by default) and it still doesn't find the files;
>>> which explains all the bug reports that we are getting that import fails
>>> for people - but due to our promiscous parser it doesn't completely
>>> fail, it's just very incomplete
>>
>>
>> I just pushed a commit that should fix this. It does two things
>> - sets .\\xslt as the path the search for XSLT files on Windows
>> - displays an error if the binary was compiled with XSLT, we identified
>>   the correct stylesheet to use but then cannot load it
>>
>> Please test
>
> that's odd, since "modern" windows should be slash agnostic.
> i've just tested and both ./xslt and .\\xslt work for me (slash at the
> end doesn't seem to matter).

I was totally baffled by it as well. So I did a make clean, make, create
new installer, uninstalled the old one, installed the new one, and it
failed. Used the backslash, same dance, and it worked.

I have absolutely no explanation. It might be that there was strong sun
spot activity. It might be that Greg KH was standing behind me and
caused bit flips in my virtual machine. But that's what I observed, so I
went with the backslash.

> so perhaps it's XSLT / cross-build related.

I'd be surprised if it was XSLT related. I'm mildly more willing to
suspect that something in the cross building makes this trip up. But
either way, the patch that I committed seems to work. Did you have a
chance to test it as well?

> it now shows the red stylesheet error, which is nice.

Thanks - the absence of an error and the promiscuity of our built in
parser was hiding this problem, that's why I added the very visible
error. I now have two reports that we may have a similar problem with
the Mac binaries, so I'll build those with the new error return and see
what I get.

But one thing this shows is that we do not have enough active testers
that test Mac and Windows. I went through the logs and when I asked for
the installers to be tested, two people each downloaded them. That just
is not enough coverage :-(

/D


More information about the subsurface mailing list