tests: salvage unit tests from meson-only unit_tests - #4281
Conversation
The tp blendmath unit test lived in unit_tests/tp, reachable only through the meson build, which no longer works. Port it to a plain C runtest (dropping the vendored greatest.h framework) that compiles test_blendmath.c together with blendmath.c at test time. Unreferenced blendmath functions are garbage-collected at link time so the whole trajectory planner does not need to be linked. Same assertions as the original test.
Salvages the G20/G21 coverage from the meson-only unit_tests/interp suite as a runtest. Verifies that switching length units converts the current position, and that the numbered parameters backing the work offsets (#5221 ff) and G92 axis offsets (#5211 ff) stay constant because they are stored in user (INI) units. The rest of the old suite duplicates existing runtests (tests/interp/g10, tests/interp/g52) and is not carried over.
The worktree has core.fileMode=false, so the chmod never reached the commit and CI failed with 'checkresult: Permission denied'.
The test compiles blendmath.c and internal TP headers from the source tree, which does not exist when testing installed packages (SYSTEM_BUILD=1). Mark it run-in-place only.
posemath.h comes from the include dir, use angle brackets. Use explicit directory prefixes in the include statements instead of adding src/emc/tp and src/emc/motion to the include path.
|
Interesting,... Now the halui/mdi test fails again. |
|
Yes, seems like its failing quite often today, almost every other build, since I merged your 4 PR's this morning |
That doesn't make sense. The PRs don't touch halui (that is still in my tree). The problem seems to be syncing halui instructions to emc via nml. |
|
BTW, the amount of CI failures has been rather high lately (disregarding sid problems) and things have been stuck many times. Not sure how much of that is our fault... But, we probably have some race issue at hand. |
|
Now all meson.build files and the unit_tests dir can be removed. |
Do you have that ready in the pipe? Or leaving the pleasure to me? |
|
Nothing in my pipeline. You are welcome to take the pleasure, if you want it. |
2 participants
First step toward #4276 (retiring the meson.build files).
The only live content depending on the meson build is
unit_tests/(806 lines). This PR salvages what is worth keeping as regular runtests:tests/blendmath: the tp blendmath unit test, ported from the greatest framework to plain C with identical assertions. It compilesblendmath.cat test time; the unreferenced rest of the trajectory planner is garbage-collected at link time.tests/interp/unit-conversion: a new runtest covering G20/G21 unit switching, the one genuine coverage gap I found. It verifies that the current position converts and that the offset parameters (#5211, #5221) stay constant, since they are stored in user (INI) units.The remainder of
unit_tests/interpis not carried over: the G10/G92/G52 cases duplicate existing runtests undertests/interp/(they were originally derived from them), two sections enshrine known-wrong rotation behavior flagged with FIXME by the original author, and the block-parse and toString tests only exercise parser internals.Verified:
scripts/runtestspasses for both new tests and the neighboring interp suites (9/9).