Hi all,
I noticed that lava-test and lava-android-test canonicalize test case IDs differently. For example, "[build] usevbo" becomes "buildusevbo" in lava-test and "build_usevbo" in lava-android-test.
This is caused by a reversal in the order of character replacement actions: lava-test first removes badchars and then replaces spaces with underscores, whereas lava-android-test does it the other way around (which is the correct way).
This issue affects tests that can run both on android and ubuntu (like glmark2), because it makes it hard to process information across test runs that ran on different platforms.
I see the following options:
1. Change lava-test behavior to match that of lava-android-test (which works as expected). This is going to create backwards-compatibility issues for some tests, but I am not sure how much people care (I don't mind this happening for glmark2).
2. Add a mechanism that can be optionally used to make the lava-test parser work correctly. For example, by making the badchars property writable one could set badchars to [^ a-zA-Z0-9._-] (note the additional space ' ' character at the beginning of the character set) to work around the problem.
3. Let the user handle this by writing a custom parser (e.g. one that would just override badchars property as noted above).
If (1) is acceptable, I would vote for it, as it will solve the problem once and for all, without the need for any action from the user.
Thoughts?
Thanks, Alexandros