113468 - URL: random urls fail to load

Status: VERIFIED WORKSFORME

(Core :: Networking, defect, P2)

Description

[Alec Flett]

ever since the new url parsing landed, I see random alerts saying that certain
directories fail to load. For instance, I just went to
http://lxr.mozilla.org/seamonkey/ and I got an alert "Unable to open
"/seamonkey/" or something very similar.

I've also noticed that certain urls fail to load stylesheets the first time, but
subsequent loads work just fine. I see this most often at CNET and
money.cnn.com... my theory is that its having a similar problem trying to load a
CSS file but we don't get alerts for stylesheets that fail to load

Unfortunately, its kind of unpredictable when it is going to happen, and after
it happens the first time, it seems to work correctly.

Comment 1

[Darin Fisher]

odd... the symptom sounds more like some sort of thread safety problem.  URL
parsing is stateless.  nsStandardURL has state, but it's usage should be
identical each time the browser loads content from the same site.  i checked in
a change to the socket transport recently that could potentially have had a bug
that would lead to strange behavior such as this.  would love to know when
exactly you started seeing this.

Comment 2

[Alec Flett]

wow, today's build (2001121103) on Win2k is really showing off this problem -
more than 50% of the images on each page are failing to load.. when I
right-click on an image, I get an alert:
"The file /foo/bar.gif cannot be found. Please check the location and try again."

strangely, I'm not seeing this on linux (2001121106)

Also, this is specifically on release builds...I'm not sure I've seen it in
debug builds.

Comment 3

[Darin Fisher]

i'll spend some time with an optimized win2k build today and see if i can't
repro the problem myself.

Comment 4

[Alec Flett]

strangely, the build now seems to be working ok.. but for the first 5-10 minutes
of usage, it was very messed up. I wonder if it might be cache-related?

My theory on why its now working is that I've visited the usual 5 or so sites
that I view each day (cnet, slashdot, etc) and with enough visits, eventually
all the images are once again in the cache. (i.e. say it messes up 20% of the
images each time - after 5 or 10 reloads, chances are you've gotten all the
images right at least once.

Comment 5

[Darin Fisher]

-> 0.9.8

Comment 6

[Darin Fisher]

-> future (adding helpwanted keyword)

Comment 7

[Darin Fisher]

marking WORKSFORME based on how old this bug is, and i believe that we've worked
out all the kinks from the URL parser rewrite of yesteryear.

Comment 8

[benc]

V/wfm.

cache sounds like a good explaination.