38496 - Viewer is no longer a useful regression testing tool

Status: RESOLVED WORKSFORME

(Core Graveyard :: Viewer App, defect, P1)

Description

[karnaze (gone)]

There are at least 2 problems with Viewer (when used with regression testing 
options), possibly due to not keeping up with the changes to systems below it. 

First, differences are flagged when there are no differences. To see this run 
"rtest baseline" in layout/html/tests/table/core and then run "rtest verify" and 
several differences will be reported.

Second, there are pages that hang (and need to have "back" and "forward" 
pressed) which have no remote images on them. There are also messages printed 
which could be related.

Comment 1

[karnaze (gone)]

Adding nsbeta2 to the keywords field. This is causing regressions in 
tables because I cannot afford to spend 6 hours each time I check in to run 
through all of my tests manually. It is probably affecting the other users as 
well.

Comment 2

[buster]

this should be a high priority bug.  this will *kill* me tomorrow when I try to 
land a ton of changes that have been accumulating in my tree for the last week.

Comment 3

[Eric Pollmann]

> First, differences are flagged when there are no differences. To see this run
> "rtest baseline" in layout/html/tests/table/core and then run "rtest verify"
> and several differences will be reported.

For cases like this that fail, I've noticed that either the baseline or the 
verify run (but not both) produces this warning.  If it's the baseline, the 
regression data is bogus and incomplete - zero size frames that should not be.  
If both produce the warning, there are no differences reported, but the 
regression data is still bogus.  I get this very frequently for iframes inside 
of tables:

0[c025e0]: WARNING: not calling OnStartRequest, file d:\builds\pollmann\mozilla\
netwerk\base\src\nsAsyncStreamListener.cpp, line 215
WARNING: not calling OnStartRequest, file d:\builds\pollmann\mozilla\netwerk\bas
e\src\nsAsyncStreamListener.cpp, line 215
0[c025e0]: WARNING: not calling OnDataAvailable, file d:\builds\pollmann\mozilla
\netwerk\base\src\nsAsyncStreamListener.cpp, line 409
WARNING: not calling OnDataAvailable, file d:\builds\pollmann\mozilla\netwerk\ba
se\src\nsAsyncStreamListener.cpp, line 409

Yes, this makes regression testing pretty much impossible without whipping out 
the pixel ruler and going through hundreds of pages one by one.

Comment 4

[Eric Pollmann]

In the failing examples I've found (all contain an iframe), 
nsWebCrawler::OnEndDocumentLoad is called twice.  The first time it is called, 
the main document is finished loading, but the iframe has not yet finished 
loading.  This is when the regression data is calculated - perhaps explaining 
the incomplete and incorrect frame data.

Should nsWebCrawler::OnEndDocumentLoad be called after the document itself 
finishes loading, or after all components of the document have loaded?

Comment 5

[Eric Pollmann]

Tossing to Rick to take a look at.  Thanks Rick!

Comment 6

[Eric Pollmann]

This works for me in tonight's build after Nisheeth's change to disable 
asynchronous reflow.  Can Chris or somebody else verify this?

Comment 7

[karnaze (gone)]

This is not fixed and as far as I can tell has been broken for almost 2 months. 
Follow the instructions as I stated in the original description.

Comment 8

[karnaze (gone)]

I should have read Eric's comments more closely and pulled Nisheeth's changes. 
It makes a lot of sense that this was causing the problem because when I ran 
viewer with a 1 seconde delay between loading pages, I didn't have a problem. 
I'll put it back as fixed (and pull the changes) and CC Nisheeth, because when 
the asynch stuff is put back, Viewer needs to be tested in this mode.

Comment 9

[karnaze (gone)]

I just updated my tree and the table regression tests are still reporting 
differences when no code changes. Eric only tried the frameset regression tests. 
Reopening.

Comment 10

[leger]

Putting on [nsbeta2+] radar for beta2 fix.

Comment 11

[rpotts (gone)]

I have no clue what is going wrong, and unfortunately I'm so swamped that I 
don't have time to wade into it right away :-(

For lack of a better person, I'm kicking this back to pollmann - sorry Eric :-(

Comment 12

[karnaze (gone)]

I'm adding "-d 1" to the regression bat files to ease the pain of this bug. 

Comment 13

[Eric Pollmann]

Chris, starting sometime either yesterday or today, I stopped having this 
problem.  I removed all the -d 1's and notice only 2 bounding box differences in  
the entire test suite.  Are you still seeing this with a pull from today?

Comment 14

[Eric Pollmann]

BTW, there are 2 asserts in content sink and a crash are present in the suite 
though, I am working on fixes for both of the asserts on bug 42734.

Comment 15

[Eric Pollmann]

Chris, handing this one back to you because I can't reproduce it any more.  Let 
me know if I can help, thanks!

Comment 16

[karnaze (gone)]

I don't have time to investigate this further. I'm pretty sure that for the 
table regression tests, the "-d 1" option helps. 

Comment 17

[Christopher Hoess (gone)]

LATER is deprecated.

Comment 18

[Christopher Hoess (gone)]

WFM per pollmann's comment.