Re: Occasional web view corruption (extra html escapes)

[Eric Wong <e@80x24.org>]

Filip Hejsek <filip.hejsek@gmail.com> wrote:
> Hello,
> 
> I have figured out why this happens.
> 
> I have reproduced the bug by doing the following:
> 1. setup an instance of public-inbox and import some message into it
> 2. create extindex named all and add it to config
> 3. start public-inbox-httpd
> 4. directly open http://<server-address>/all/<msg-id>/

OK, but I wasn't able to reproduce it on my setup since I
had coderepos configured :x

Coderepos were hiding the bug from me.

> The server will enter the broken state if this is the first page loaded
> from the server.
> 
> The issue occurs because of the following sequence of events:
> 1. WWW->call is called
> 2. after matching the URL, msg_page is called
> 3. msg_page validates the inbox name by calling invalid_inbox_mid
> 4. invalid_inbox_mid calls invalid_inbox
> 5. the name is looked up with lookup_name
> 6. because there is no inbox with that name, undef is returned
> 7. the name is looked uo with lookup_ei
> 8. the lookup succeeds
> 9. get_mid_html is called to generate the HTML page
> 10. addr2urlmap is used to construct a regex of known addresses
> 11. because no inbox has been instantiated, $cfg->{-by_addr} is empty
> 12. because of that, $re will an empty string
> 13. so the final regex is /\b()\b/, which matches every word boundary

Thanks for the diagnosis and excellent explanation!

The reason I didn't hit this problem on my instance was because
configuring coderepos causes CodeSearch->load_coderepos to call
lookup_eidx_key, which calls _lookup_fill and populates {-by_addr}

So I think adding a guard to prevent using regexp from empty -by_addr
is a first step; but additional checks + preload fixes should
probably be needed...

Fixes coming...