[feat] Associate a unique hash with each test #2578
Conversation
|
@flx42 Could you test this PR with your scenario? Now filelog performance files should not change across reframe invocations. |
Codecov Report
@@ Coverage Diff @@
## master #2578 +/- ##
==========================================
- Coverage 86.69% 86.68% -0.02%
==========================================
Files 59 59
Lines 10827 10854 +27
==========================================
+ Hits 9387 9409 +22
- Misses 1440 1445 +5
Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here. |
Actually, 4.x does not change how the I would see this as a separate feature request, especially if the current behaviour breaks your workflow or just annoys you 😂 Technically, it is possible, but it might need to add another configuration parameter to the |
|
Sounds good! Then I will file a separate feature request at a later time
when I'm back in the office as I'm currently travelling.
Thanks!
…On Tue, Aug 30, 2022, 10:45 Vasileios Karakasis ***@***.***> wrote:
But it doesn't solve the part about being able to customize the perflog
output file to get a behavior similar to 3.x if you want to, not sure what
is your opinion on this.
Actually, 4.x does not change how the filelog handler generates the log
filenames. In the filelog handler you could only (and still can)
configure the prefix directory of the log file, but not the name of the
filename. The name of the filename was composed by the test's unique_name,
thus the ambiguity with the variant names, which is now solved with hashes.
In the old naming scheme, the unique_name contained a stringification of
the test parameters and their values, which is essentially the same as of
now, except that the string representation is now different.
I would see this request as a separate feature request, especially if the
current behaviour breaks your workflow or just annoys you 😂.
Technically, it is possible, but it might need to add another configuration
parameter to the filelog handler to control the log file name per se.
—
Reply to this email directly, view it on GitHub
<#2578 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AA32BDKFLILQI22BVXKKMNTV3XCZFANCNFSM567UDBIA>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
|
4 participants
Each test is now associated with a hash code, which is generated by the test name and the key value pairs of its parameters. This hash code is printed along with the display name with the
-loption and tests can be selected also by their hash code with the-noption, by passing-n /<hash>. Tests do still have their unique name that is comprised by the class name and the variant number. Test-specific files associated with the test execution, such as the stage and output directories and the performance log file of thefileloghandler use the test's hash code in their filename. Conversely to the variant number, the test's hash code does not change if new values are added to a parameter, or values are deleted or the list of values are shuffled. However, you might get filesystem conflicts if you run a parameterised test that has a parameter that takes the same value many times. Such tests must run serially to avoid this. Note that if a parameter is added to or deleted from the test or the name of parameter changes, the hash will change.Fixes #2577.