1357134 - Verify win64 TC-derived builds are equivalent to BB builds

Status: RESOLVED FIXED

(Firefox Build System :: Task Configuration, task)

Comment 1

[Chris Pearce [:cpearce (Not reading bugmail)]]

Attachment: Bug 1357134 - Recover from incorrectly guessing the CDM's shmem sizes.

We are pre-allocating shmems in the content process for use by the CDM in the
GMP process. We guess the size of shmem required. However if we guess wrong,
currently we always end up taking the non-shmem path for video frames to
return to the content process, which results in us sending another shmem
(of the wrong size) to the CDM, and this continues until we hit the limit
on the number of shmems that we tolerate the CDM asking for.

So in this patch, I change our behaviour to detect when we're allocating
shmems that are too small, whereupon we purge the existing shmems and switch
to allocating them at the size being requested by the CDM.

This means we recover from incorrectly guessing the size of shmems required
by the CDM. The overhead of an incorrect guess should be one video frame
transferred via the nsTArray path.

Review commit: https://reviewboard.mozilla.org/r/134568/diff/#index_header
See other reviews: https://reviewboard.mozilla.org/r/134568/

Comment 2

[Chris Pearce [:cpearce (Not reading bugmail)]]

Comment on attachment 8862704 [details]
Bug 1357134 - Recover from incorrectly guessing the CDM's shmem sizes.

Wrong bug...

Comment 3

[Kim Moir [:kmoir] ET]

kmoir> Kim Moir callek: do you have instructions on how you verified the linux builds before that I can use to build upon for the windows work?
<Callek> Justin Wood kmoir: my best docs are in 1321005 and 1320998
12:18 PM 
<kmoir> Kim Moir okay will look thanks

Comment 4

[Kim Moir [:kmoir] ET]

So I tried installing diffoscope locally but it failed due to some dependency issues, then I tried in a virtualenv same issues.  The online compare tool failed with an error

https://try.diffoscope.org/zrcdqezffegz

Tomorrow I will try to install it on a loaner and compare the file there.

Comment 5

[Kim Moir [:kmoir] ET]

I've tried many incantations to run diffoscope on a virtual env on my mac, and the same on a loaner linux machine. None of them work. I talked to Callek and ran the same revision of the diffoscope tool that he used and the same libs installed in the virtual env and still it doesn't work.

So Callek since you have this tool working, could you run it with the following files and send me the output to analyze

for win64

old  (m-c)
https://treeherder.mozilla.org/#/jobs?repo=mozilla-central&filter-job_type_symbol=N&selectedJob=115169646
https://archive.mozilla.org/pub/firefox/nightly/2017/07/2017-07-18-03-02-07-mozilla-central/firefox-56.0a1.en-US.win64.zip

new (date) 
from https://treeherder.mozilla.org/#/jobs?repo=date&revision=edbab709cca954d79a2c227d4e77d8714cb9312f&selectedJob=115258897
https://queue.taskcluster.net/v1/task/ECGfLKQNQy6TXV7S7SEaPQ/runs/0/artifacts/public/build/target.zip

diffoscope --new-file --text /Users/kmoir/diff_win64.txt  ~/verify64/firefox-56.0a1.en-US.win64.zip ~/verify64/target.zip

Comment 6

[Kim Moir [:kmoir] ET]

I have the tool working and am comparing the files.

Rail created a docker image called rail/diffoscope with the tool which I was able to download and use

The issue with my installation was that the pip packages were broken. The debian packages work.  

I haven't found any unexpected issues so far other than the fact that the packages are from different revisions on m-c vs date.  Callek said he would provide me with a new revision on date that I could compare more accurately with m-c.

Comment 7

[Justin Wood (:Callek)]

Task for win64 repackage signing (Installers + mar): https://tools.taskcluster.net/task-inspector/#fAOfyNNLRuu4TCsKzOWwWg
Task for win64 signing (.zip): https://tools.taskcluster.net/task-inspector/#RSke61aYR32XYDvQsSnk-A
Corresponding BB Nightly build (files): https://archive.mozilla.org/pub/firefox/nightly/2017/07/2017-07-18-09-40-46-mozilla-central/

Comment 8

[Kim Moir [:kmoir] ET]

I've been comparing the files from the comment above and haven't found anything yet other than update channel differences as expected.

Comment 9

[Kim Moir [:kmoir] ET]

verification by PI team is complete and found no tc specific issues