Re: Occasional web view corruption (extra html escapes)

[Konstantin Ryabitsev <konstantin@linuxfoundation.org>]

On Wed, Sep 04, 2024 at 01:14:43PM GMT, Eric Wong wrote:
> OK, and nothing malformed or commented out in the address=
> fields?  And no stray semi-colons or hash marks in any address
> fields which comment out only the value?

Nope.

> Because having an empty address= and empty url= field in my
> publicinbox.lkml section gives me something close (but not
> exactly) to what you have:
> 
> <a
> href="/?t=20240829184845"></a> &<a
> href="/?t=20240829184845"></a>lt<a
> href="/?t=20240829184845"></a>;<a
> 
> Note the href has a leading slash in my near-reproduction
> but yours did not.
> 
> It's not a double-escaping problem, but the substituion is
> is breaking already -escaped `&lt;' and `&gt;' apart with <a>
> tags in between characters.  but I'm not sure how this is
> happening to you if all your address fields look OK.
> 
> I'll filter out blank addresses in our config reader, but
> I also wonder if there's anything else going on...

I think so, because the timestamp I have on the config file is Aug 20, and the
corruption manifested itself on Sep 2 or 3. Previously this happened on Aug
26, which is the last time I would have restarted the public-inbox-httpd
process. The only notable thing I currently see about these events is that
they were on a Monday 1 week apart, but it could have been an accident.

> And all your writes to the config are via git-config?

Yes, they are written as part of public-inbox-init. This is the chunk of code
where that happens:
https://git.kernel.org/pub/scm/utils/grokmirror/grokmirror.git/tree/grokmirror/pi_indexer.py#n164

> Fwiw, the buggy code would be in addr2urlmap called by
> _msg_page_prepare in PublicInbox::View.
> addr2urlmap() will escape any regexp metacharacters present in
> addresses via quotemeta, so there's no chance of regexp
> injection from a config.
> 
> (hopefully coherent, running on fumes due to real life things)

One thing I can do next time this happens is take that node out of rotation
and keep it in that state for additional debugging.

Thanks for your help!

-K