Replace parallel statements with Caffeine calls #136
Conversation
This replaces the `sync all` statement in program_termination_s with call prif_sync_all so that we use the prif/caffeine capability rather than mistakenly invoking a compiler's pre-existing parallel statement support if it exists.
This replaces the `error stop` statement in example/hello.f90 with `call prif_error_stop` so that we use the prif/caffeine capability rather than mistakenly invoking a compiler's pre-existing parallel statement support if it exists.
This commit 1. Replaces all occurrences of `error stop` with `call prif_error_stop` in the programs launched by unit tests for program termination and 2. Expands the number of such programs so that one each program each tests the six cases of normal and error termination with integer stop code, character stop code, and no stop code. 3. Adds unit_test_parameters_m to put expected stop codes in one place for use by unit tests, including the programs in example/support-test that support unit tess in test/.
Co-authored-by: Katherine Rasmussen <krasmussen@lbl.gov>
This commit fixes some of the example/support-test program names to match the base file names and adds cmdstat and cmd_msg arguments to the execute_command_line invocations in the test suite. The latter change should facilitate capturing richer information about error conditions resulting form launching a child process to run one of the programs employed for testing normal and error termination.
There was a problem hiding this comment.
Noted a few minor but important changes I'd like to see.
Not represented in my review is the fact I consider the entire approach taken in the exit testing of invoking fpm from within a unit test and inspecting the exit code to be very fragile. I've just entered this as issue #137
Co-authored-by: Dan Bonachea <dobonachea@lbl.gov>
This will allow a human to audit that the string appears in the output when run manually. Co-authored-by: Dan Bonachea <dobonachea@lbl.gov>
This will allow a human to audit that the provided string appears in the output. Co-authored-by: Dan Bonachea <dobonachea@lbl.gov>
There was a problem hiding this comment.
Thanks for merging the changes. IMO overall this PR is definitely an improvement, and I'd like to see it in the release.
I've identified one other serious defect in the exit tests modified by this PR, which are not working as intended even in shared memory. The same defect also exists in main, so this is not technically a regression.
Given the impending release and the fundamental problems documented in #137 that make this entire approach to exit-testing a non-starter in general, fixing this particular defect is basically "putting lipstick on a pig". So I'm also fine ignoring the problem for now and merging as-is.
Co-authored-by: Dan Bonachea <dobonachea@lbl.gov>
This PR
error stopandsync allwith theprif_error_stopandprof_sync_allcalls, respectively.prif_error_stopwith no stop code.unit_test_parameters_mmodule with constants for use in different parts of the test suite.