672538 - remotereftest.py and runtestsremote.py reference dm_auto before initialized

Status: RESOLVED FIXED

(Testing :: General, defect)

Description

[Geoff Brown [:gbrown]]

If I execute: MOZ_HOST_BIN="../objdir-firefox/dist/bin/" make reftest-remote
I get a failure:

Traceback (most recent call last):
  File "_tests/reftest/remotereftest.py", line 447, in <module>
package set: org.mozilla.fennec_unofficial
    main()
  File "_tests/reftest/remotereftest.py", line 396, in main
    dm = dm_auto
NameError: global name 'dm_auto' is not defined
/bin/sh: @errors=: not found
reftest-remote passed

Similarly, if I execute: MOZ_HOST_BIN="../objdir-firefox/dist/bin/" make mochitest-remote

I get:
Traceback (most recent call last):
  File "_tests/testing/mochitest/runtestsremote.py", line 351, in <module>
    main()
  File "_tests/testing/mochitest/runtestsremote.py", line 318, in main
    dm = dm_auto
NameError: global name 'dm_auto' is not defined
make: *** [mochitest-remote] Error 1

Comment 1

[Geoff Brown [:gbrown]]

I also found that some directories were not being created. For example, if <root>/tests does not exist at the start of a mochitest, the profile directory is not created, and all tests fail. This is caused by a recent change to devicemanagerADB.pushDir (that I put in!); simple fix on the way...

Comment 2

[Geoff Brown [:gbrown]]

Attachment: patch for review

With these simple changes I can successfully run:

cd objdir-droid
MOZ_HOST_BIN="../objdir-firefox/dist/bin/" make reftest-remote
MOZ_HOST_BIN="../objdir-firefox/dist/bin/" make mochitest-remote

(devroot of /mnt/sdcard/tests, with devroot non-existent before test; not all tests pass, but the test suites appear to run to completion).

Comment 3

[Geoff Brown [:gbrown]]

I just realized/recalled that the dm_auto vs dm_none issue is covered by jmaher's patch at https://bug669549.bugzilla.mozilla.org/attachment.cgi?id=545702, which never landed. Was that an oversight? Do we also want to add the optional run-as package setting now?

Comment 4

[Brad Lassey [:blassey] (use needinfo?)]

(In reply to comment #3)
>  Do we also want to add the optional run-as
> package setting now?

Yes and no. I think we want to change that to an optional "run as root" flag so we can test release builds on rooted phones

Comment 5

[Joel Maher ( :jmaher ) (UTC-4) (on leave, back November 5th)]

well, we cannot specify org.mozilla.fennec with a nightly build, so that patch is not very useful.  I think if we have packageName=org.mozilla.fennec, we shouldn't use it, only use the packageName of org.mozilla.fennec_unofficial.

Comment 6

[Joel Maher ( :jmaher ) (UTC-4) (on leave, back November 5th)]

Comment on attachment 546828 [details] [diff] [review]
patch for review

Review of attachment 546828 [details] [diff] [review]:
-----------------------------------------------------------------

looks good assuming the "/x" is intended.

::: build/mobile/devicemanagerADB.py
@@ +96,5 @@
>      # contains symbolic links, the links are pushed, rather than the linked
>      # files; we push file-by-file to get around this limitation
>      try:
> +      if (not self.dirExists(remoteDir)):
> +        self.mkDirs(remoteDir+"/x")

why the +"/x" here?

Comment 7

[Geoff Brown [:gbrown]]

devicemanager.mkDirs takes a *filename* as a parameter, and creates all the directories in the path to that filename. So if you want to create directory /d1/d2/d3, you need to mkDirs("/d1/d2/d3/something"). (I don't particularly approve of the interface, but that is the way I found it!)

Comment 8

[Geoff Brown [:gbrown]]

(In reply to comment #5)
Let's defer the packageName-setting/run-as/root-access issues to another bug, if/when necessary, and check in this patch as it is.

Comment 9

[Joel Maher ( :jmaher ) (UTC-4) (on leave, back November 5th)]

agreed.  The more people that can use this the better.

Comment 10

[Marco Bonardo [::mak]]

http://hg.mozilla.org/mozilla-central/rev/ff75a2a2e771