Ingest coverage data only for the files in the filter that actually have some

[dvdoug]

For #876. The Psalm failure is bogus, its not even on a changed line!

[codecov]

Codecov Report

Merging #877 (1ae01b6) into 9.2 (3ca2f86) will decrease coverage by 0.06%.
The diff coverage is 0.00%.

@@             Coverage Diff              @@
##                9.2     #877      +/-   ##
============================================
- Coverage     83.10%   83.03%   -0.07%     
- Complexity     1116     1117       +1     
============================================
  Files            62       62              
  Lines          3652     3655       +3     
============================================
  Hits           3035     3035              
- Misses          617      620       +3     

Impacted Files	Coverage Δ
src/Driver/PcovDriver.php	44.44% <0.00%> (-8.89%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 3ca2f86...1ae01b6. Read the comment docs.

[sebastianbergmann]

@krakjoe You wanted a ping when the PCOV driver is changed.

[sebastianbergmann]

@indreka Can you please test whether this patch improved PCOV driver performance for you? Thanks!

[krakjoe]

There is no sense in entering collect() or calling clear(), if there is nothing returned from waiting()

[indreka]

@sebastianbergmann I can confirm that the proposed change works very well!
Output of coverage.xml is same when I compared my originally proposed "hack" vs proposed solution.
In addition the speed has also improved, subset of testsuite with my "hack" locally runs 55sec, the proposed solution makes it run with 35sec! Will add custom patching to circleci runner also to verify final speedup.

[sebastianbergmann]

There is no sense in entering collect() or calling clear(), if there is nothing returned from waiting()

Thank you for your feedback. However, I am not sure whether the patch changes the code to do what you propose (in which I case I would accept it). Might be because I have a hard time understanding the line of code in question due the use of the ternary operator.

@dvdoug Could you please change the code to not use a ternary? Thanks!

[krakjoe]

However, I am not sure whether the patch changes the code to do what you propose

I'm not sure if that was directed at me.

For the avoidance of any doubt: Do not call collect unconditionally. Do not call clear if nothing was returned from waiting().

This is for performance reasons, and this code is extremely perf sensitive; It cannot make unnecessary calls.

[sebastianbergmann]

I'm not sure if that was directed at me.

It was, sorry for not being clear about this.

For the avoidance of any doubt: Do not call collect unconditionally. Do not call clear if nothing was returned from waiting().

This is for performance reasons, and this code is extremely perf sensitive; It cannot make unnecessary calls.

@dvdoug Can you incorporate Joe's suggestions? Thanks!

[indreka]

Can confirm that in circleci this change seems to make coverage scan ~5% faster when comparing to previous phpunit 8.5 + pcov setup we had

[dvdoug]

For the avoidance of any doubt: Do not call collect unconditionally. Do not call clear if nothing was returned from waiting().

Happy to reorganise things to do this, but if a testcase runs without managing to cover a single line of relevant source code then I'd honestly be pretty surprised. I'd have thought that the overhead of including an additional if here in userland would be slower than an occasional extra call into the extension during a zero-coverage scenario.

[sebastianbergmann]

Thank you, @dvdoug!