Faster test suite #243
Faster test suite #243
Conversation
git-svn-id: http://caml.inria.fr/svn/ocaml/branches/faster-testsuite@16437 f963ae5c-01c2-4b8c-9fe0-0dff7051ff02
git-svn-id: http://caml.inria.fr/svn/ocaml/branches/faster-testsuite@16439 f963ae5c-01c2-4b8c-9fe0-0dff7051ff02
For some tests, we reduce the size of the data sets. For some other tests that started as benchmarks, we avoid repeating the test needlessly. git-svn-id: http://caml.inria.fr/svn/ocaml/branches/faster-testsuite@16440 f963ae5c-01c2-4b8c-9fe0-0dff7051ff02
|
On my machine, the testsuite before this patch takes about 3m30s to run (when already hot), while after the patch it only takes 2m20s. For reference, (Eventually we'll want to be able to run some of the tests in parallel to further gain speed on beefy machines, but this requires a more invasive redesign of the testsuite scripts. |
Not tested under Windows. Some tests are still timing-sensitive and could fail if the test machine is under very heavy load. git-svn-id: http://caml.inria.fr/svn/ocaml/branches/faster-testsuite@16446 f963ae5c-01c2-4b8c-9fe0-0dff7051ff02
git-svn-id: http://caml.inria.fr/svn/ocaml/branches/faster-testsuite@16447 f963ae5c-01c2-4b8c-9fe0-0dff7051ff02
|
The latest commits revise tests/lib-threads so that it takes less time (30 secs instead of 1 minute) and (one can hope) is more reliable (less sensitive to timing effects). Not tested under Windows, however. |
|
fyi, I was also working on the test suite, in particular I want to support cross-compile targets. The idea is to compile all tests into a single binary that is started as a TCP server on the device. The compilation and scripting still takes place on the host. So far it mostly works by providing additional Makefile targets. It's not yet finished, but one of the problems is that there are too many special tests (i.e. neither using Makefile.several or Makefile.one). If that could be improved this would be great. Find my work here: https://github.com/gerdstolpmann/ocaml/tree/gs-4.02.3%2Bios%2Btestsuite/testsuite |
|
@xavierleroy , do you plan to merge this in trunk eventually? I've been doing my stuff-to-merge testing on top of this branch, and it would be convenient to have this directly in trunk because there are small conflicts ( against the french comment translation work, which changed stuff in If you think the current state is mergeable I can do it this week-end, with the other easy PRs. |
|
Tested on MSVC-32 and mingw-64. OK for merging. |
|
Merged (rev 16466). |
doc changes for wait_for and wait_until
We were generating a map of code ids to code by traversing the unit, but the only caller of `compute_offsets` already has an `Exported_code.t` at hand, so we can just use that and avoid a double traversal.
We were generating a map of code ids to code by traversing the unit, but the only caller of `compute_offsets` already has an `Exported_code.t` at hand, so we can just use that and avoid a double traversal.
We were generating a map of code ids to code by traversing the unit, but the only caller of `compute_offsets` already has an `Exported_code.t` at hand, so we can just use that and avoid a double traversal.
We were generating a map of code ids to code by traversing the unit, but the only caller of `compute_offsets` already has an `Exported_code.t` at hand, so we can just use that and avoid a double traversal.
The test suite takes significant time to run. This is annoying when testing on small devices (Raspberry Pi) or, even worse, using emulators (QEMU).
This pull request reduces the running time of the test suite by
With the proposed changes, almost every test run in 1 second or less (in bytecode as well as in native code) on a modern PC. The exception is tests/lib-threads, which still takes a long time (about 1 minute), and needs to be redesigned and rewritten.