Re: [documancer] broken pipe problem

[<arn...@we...>]

[...]
> > with the following I can produce a broken pipe
> > (see the traceback at the end):
> >   - Book: Python 2.3 Docs
> >   - Search String: AF_UNIX
> >   - Results:
> >     a) 1000 socket
> >     b)  889 7.2.1 Socket Objects
> >     c)  889 Index
> >     d)  873 11.15 SocketServer
> >   - wait until page a) is displayed
> >   - click on d), wait until it is displayed, everything is ok
> >   - click on c), do _not_ wait until it is displayed (it takes
> >     a little while because it is a long text)
> >     and click e.g. on a) ==> "broken pipe"
>
> This is harmless (the exception is catched, it only polutes stderr),
> but thanks anyway ;) It's now fixed, Documancer's HTTP server now
> handles all (HTTP) errors silently.

I am not quite sure if it is harmless - at least doing
this ("clicking on a search result before it is displayed")
a couple of times  I have the effect that documancer
does not react on clicks in the search results list
or on changing a book.
(Sorry, I should have added this description after the "broken pipe"
in the first place...)

> are there any  other characters that can be part of identifier?

Hmm, maybe this depends on the context, what about "-"?
(but I haven't really thought about this. I think
one could leave it as is at the moment and change
the behaviour once users complain. If it really
should turn out to be dependent on the documents
which are searched one could think of adding this as a configuration
option for the corresponding book. But at this point
that would be featurism IMHO ...;-)

> Note that you have
> to reindex the book for the fix to kick in.

Works perfect - many thanks.

> > - Even type-ahead find works to some extent,
> >   presently only for the text of links (!?).
> >   Do you think this is just a wxMozilla configuration issue ?
>
> Yes, it is -- set "accessibility.typeaheadfind.linksonly" to "false"
> in prefs.js.

Great, this works fantastic, many thanks.

Best, Arnd