[tclwebtest] Re: tclwebtest slowness

[Tilmann S. <ti...@ti...>]

[replying to the tclwebtest list where this belongs to]

* Peter Marklund <pet...@fa...> [20031022 13:01]:
> Turns out there is a bit of a slowness problem with Tclwebtest on the
> /acs-lang page. Why? Maybe because of a huge select box with timezones.
> Below is a snippet of debug printouts. The parse_caption_before <number>
> statements pop out *really* slow, as in once a second or so. Do you have
> any idea why this is so?
> 
> I've tracked down that it's when I invoke form find on the page that the
> slowness begins.
> 
> Please find attached the acs-lang index page.
> 
> following a redirect to:
> http://dotlrn20-pg-test.collaboraid.net/acs-lang/
[...]

I don't have an answer for this specific problem, just the general
observation that the way tclwebtest parses html is bound to be slow
and potentially very slow with big pages or unexpected html
structures. I did some experiments with the html parsing mode of tdom
and some other html parsing library which was written in tcl but found
both not sufficient for various reasons. The main problem with tdom
remains (last tested version 0.7.7) that it is not forgiving enough
for malformed html.

Last time I tested against openacs html it borked on the site-map, on
some pages in bug-tracker and propably on some parts of the standard
form templates as well.

Of course changing to a more strict html parsing mode would produce a
stronger motivation to fix those very broken html pages, which would
be good for openacs, but would narrow the usability of tclwebtest.

Also tdom would add an installation requirement.

I'd rather use an html parsing library that is almost as lenient as
tclwebtest is right now, but I didn't find it yet.


That all said, it might well be that the specific slowness you are
talking about can be fixed by tweaking some regexes, but I don't have
time to look into it right now.

cheers, til


-- 
http://tsinger.com