[sldev] SNOW-222 : request before committing the patch

[Philippe (Merov) Bossut]

Hi,

I spent the best part of the afternoon tracking that vexing Mac problem and,
as expected, that's a silly issue: the export script didn't export the new
media UI textures. Those being loaded at launch would hose the texture
loading thread entirely. Dropping the missing textures in the
skins/default/textures folder solve the problem. I attached a zip file
containing the missing textures in the JIRA. Snowglobe Mac devs, thanks in
advance for your test and confirmation (or not) of the fix.

This bug however points to another more concerning issue: when files fail to
load, the file descriptors don't seem to be released and/or the fetch
workers don't close properly. I haven't identified which yet but that's
something that could explain quite a bit of spurious "gray texture" issues
that have been reported here and there.

Going to bed now :)

Cheers,
- Merov

On Fri, Sep 25, 2009 at 6:50 PM, Philippe (Merov) Bossut <
merov at lindenlab.com> wrote:

> Hi guys,
>
> I've been banging my head all day on a vexing problem with this patch on
> Mac. Aimee and me reported the same issue on IRC but we both wrote it off on
> some silly install we did on our dev environment. Seemed weird enough we'd
> get the same issue so digging deeper, it looks like the problem is more
> serious. I won't go into the gory details but here's the short version:
> - On Mac, you get to a point where no texture download at all, you
> basically walk in a land of gray...
> - It's not systematic mind you (things do sometimes work fine)
> - When it starts though, it's there to stay
> - the trunk version or plugin-api are not affected: so it's something I did
> when merging (haven't identified what yet...)
> - Clearing the cache doesn't work: actually, it's a way to get the problem
> to repro. It seems that the file texture.cache can't get created at all...
> - Launching an unaffected version of the viewer makes the problem go away
> for a while... till it starts again...
>
> From the logs and some tracking, it seems that, at some point, we just
> can't open a file anymore. Could happen even before we even try any neywork
> activity (UI bitmaps for instance not loading in the login screen). I'm
> wondering if the updated libs used with the patch (see install.xml) could be
> the cause of this...
>
> Not great news I know but I wanted to give some update before the weekend.
>
> Cheers,
> - Merov
>
>
> On Thu, Sep 24, 2009 at 6:05 PM, Philippe (Merov) Bossut <
> merov at lindenlab.com> wrote:
>
>> Hi guys,
>>
>> I'm asking for help again: Pixel Gausman found a problem with running on
>> Windows XP SP2 if the path to SLPlugin.exe (the external plugin launcher)
>> contains white space (see PJIRA for details). I cannot reproduced the
>> problem on Windows XP SP3. Could someone else try to confirm the problem?
>> Other flavor of Windows? Thanks...
>>
>> Also, though the code path is pretty different between Windows and
>> Linux/Mac wrt to process launching, it's not a bad idea to test with various
>> exotic path... (spaces, diacritic characters, non latin...).
>>
>> Cheers,
>> - Merov
>>
>>
>> On Wed, Sep 23, 2009 at 10:57 AM, Philippe (Merov) Bossut <
>> merov at lindenlab.com> wrote:
>>
>>> Hi guys,
>>>
>>> So, there's this big monster of a patch I posted on PJIRA since a week
>>> now. Thanks a millions to those who tested it and reported issues
>>> (incorporating fixes now). In the meantime, I've been working on a "Hello
>>> World" plugin example so that I can contribute to a first draft of a "How to
>>> write a plugin" doc and get some of you guys to start hacking.
>>>
>>> I'm happy to report that I do have such a plugin working on my machine
>>> (yeah!) and, of course, that got me some new ideas on how to improve things,
>>> develop new plug ins, extend the plugin architecture, etc... all things I
>>> can't do while the big patch hasn't been committed.
>>>
>>> Add to that that I need to svn update my local repo and tweak that patch
>>> every time a commit is done on trunk and you'll understand why I'm eager to
>>> see this landing.
>>>
>>> So, what else do we need to commit this? I heard that we need more tests
>>> but it seems we don't get any significant amount of tests unless we land in
>>> trunk and produce binaries... which really makes me think I should commit
>>> this before Friday so we can make the 1.2 schedule Robla emailed earlier.
>>> What do you guys think? Is there one think that we need to do before that
>>> first commit?
>>>
>>> One last note : the bulk of this code has been tested and qualified
>>> internally already so it's not completely untested code. That makes me a
>>> little more confident with the process.
>>>
>>> Cheers,
>>> - Merov
>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.secondlife.com/pipermail/sldev/attachments/20090927/6304a311/attachment.htm