Skip to content

NEWS.md

Latest commit

 

History

History
304 lines (192 loc) · 9.54 KB

NEWS.md

File metadata and controls

304 lines (192 loc) · 9.54 KB
Raw

Version (development version)

Bug Fixes

  • check() failed to set the future plan when the specified plan contained the word "plan", e.g. --test-plan=planned::foo.

  • subset_tests(args, defaults) ignored the arguments of the tests.

  • along_test_plans() dropped errors and NULL results. Errors are now returned as the result for the plan, and evaluation continues.

  • check_plan(defaults = list()) silently skipped tests without arguments.

  • run_tests(defaults = NULL) failed with an error.

  • make_test() accepted a missing title (NA).

  • check(timeout = <seconds>) no longer leaves option future.tests.timeout set after returning.

  • check() ignored command-line options --session-info and --debug.

Test Updates

  • Now asserting that immediate conditions are relayed, where possible.

  • The test for cancel() and value() now asserts that value() on a canceled future signals a FutureCanceledError, which inherits from FutureInterruptError, also when canceled with interrupt = FALSE, when the backend does not support interrupts, and for lazy futures. A new test asserts the same for a future that has already been collected. The new assertions are skipped when testing against a version of the future package that does not yet include the corresponding bug fix.

  • Add tests for value() on lists of futures asserting that (i) argument init of reduce, and arguments stdout and signal, are respected also with inorder = FALSE, (ii) an error cancels and resolves the remaining futures, without launching lazy ones, (iii) a future that returns an error object is not mistaken for a failed future, and (iv) standard output and messages of all futures are relayed in order also with drop = TRUE.

  • Add test asserting that futureCall() passes language objects, e.g. symbols and calls, in argument args as-is to FUN.

  • The above tests are skipped when testing against a version of the future package that does not yet include the corresponding bug fix.

  • Add tests for (i) globals specified as a character vector with attributes add and ignore, (ii) unused variables not being exported as globals, (iii) a global ... together with a global ...future.FUN(), (iv) switching plan while a future is running, (v) repeated value() on an interrupted future, (vi) launching many futures without leaving unused connections behind, (vii) immediate conditions not being re-signaled, and (viii) makeClusterFuture() passing language objects as-is and supporting clusterEvalQ() that only attaches packages.

  • Add tests for (i) future(..., packages = "missing-package") signaling an error, (ii) value(<listenv>, idxs), (iii) value(<list>, reduce = structure("+", init = 42)), (iv) result() on a canceled and interrupted future, and (v) errors from a makeClusterFuture() cluster being signaled as by other clusters.

Bug Fixes

  • check() produced garbled output with interleaved spinner frames and test titles when standard output or standard error was not a terminal, e.g. when the output was piped or written to a log file. The in-place progress is now only shown when both are terminals, which can be overridden by option future.tests.progress.

Miscellaneous

  • Package utils is now declared as an imported package instead of just a suggested package.

Version 1.0.0

Significant Changes

  • Changed the package license to permissive Apache License (>= 2).

New Features

  • Add logical R option future.tests.devel, which can be set by environment variable R_FUTURE_TESTS_DEVEL. This controls whether tests with a "devel" ("under development") tag will be included.

  • check() reports on future version.

Test Updates

  • Add tests for cancel().

  • Assert that plan() accepts argument interrupts.

  • Remove tests on future(..., earlySignal = FALSE/TRUE), because that is about to be deprecated in the future package.

  • Relax assertions in test of resolved() with lazy futures.

Bug Fixes

  • Tests on plan() adjusting argument workers were not skipped if the plan tested already sets argument workers to a fixed value, e.g. plan(cluster, workers = cl).

Version 0.9.0 [2025-04-14]

New Features

  • Now the test code is displayed if there is a check error.

Version 0.8.0 [2024-03-31]

New Features

  • Now load_tests() returns a named list, where the names corresponds to the test titles.

  • Now future.tests::check("mirai::daemons(2); plan(future.mirai::mirai_cluster)") works.

Test Updates

  • Remove plan() test on legacy constant backend.

Version 0.7.0 [2023-05-20]

Bug Fixes

  • Tests asserting correctness of nbrOfWorkers() could produce an "invalid format '%d'; use format %f, %e, %g or %a for numeric objects" error when trying to produce an assertion error on nbrOfWorkers() having an incorrect value. This could happen if nbrOfWorkers() returned a non-integer value, e.g. +Inf.

  • Test asserting that the workers argument can be a function would not always work if testing with a hardcoded number of workers according to plan().

  • Test asserting that the workers argument can be a function would not work if the backend's default value was non-numeric, e.g. the cluster backend defaults to the character vector parallelly::availableWorkers().

  • Test asserting that lazy futures would be automatically launched and resolved relied on a legacy version of the Future API, where calling resolved() on a lazy future could leave it in a lazy state, which is no longer correct. A lazy future will always be launched if one calls resolved() on it.

  • Test asserting that the ff package worked across multiple futures assumed that the package is loaded automatically by the future, which it is not. The could cause the test to fail for some future backends.

Version 0.6.0 [2023-03-11]

New Features

  • Now tests can be formally skipped by calling future.tests::skip_test() from within the test. Skipped tests are counted and reported in the summary.

  • Now check_plan() outputs the reason for a test being skipped.

  • Now check_plan() outputs also the error message, error class, the call, and any standard output, whenever there is a test error.

  • Now check_plan() outputs also the test iteration index.

  • Added Rscript -e future.tests::check --version.

Bug Fixes

  • Some tests assume that the future strategy tested has a workers argument, which is not true for all future backends. For example, 'sequential' does not take argument workers. Previously, we avoided this problem by only testing if the evaluator inherited multiprocess, but that is not sufficient, e.g. upcoming future.redis::redis inherits multiprocess, but still does not have a workers argument. Now we check for the workers argument instead.

Version 0.5.0 [2022-12-15]

Tests Updates

  • Assert that none of future(), run(), result() and value() update the RNG state.

Version 0.4.0 [2022-11-21]

Tests Updates

  • Assert that rm(a) in a future expression only removes a local variable a, but never a global variable a.

  • Assert that packages data.table and ff are not affected when a future resets the R options on the worker when resolved.

  • Assert that a global that is a copy of a non-exported package object (e.g. utils:::str2logical()) is not dropped because it belongs to a package namespace.

  • Assert that ... can be exported as a global to a future, and used as-is inside a function that does not have ... arguments.

New Features

  • check() and check_plan() gained argument local, which is passed down to run_test().

  • check() gained argument envir, which is passed down to run_test().

Bug Fixes

  • The TestResult class did not record whether the test was evaluated in a local environment or not.

  • A too strict internal assertion would give Error in evaluate_expr(test$expr, envir = envir, local = FALSE, output = output, : identical(Sys.getenv(), old$envvars) is not TRUE for R 4.2.x and R-devel on MS Windows. This was because it is not possible to remove environment variables on MS Windows; they can only be set to an empty value.

Version 0.3.0 [2021-10-09]

Tests Updates

  • Assert that future() doesn't change the RNG kind.

  • Assert that future(..., conditions = character(0L)) muffles all conditions.

New Features

  • For robustness, using explicit stringsAsFactors = FALSE internally.

  • evaluate_expr(), which is used for running all tests, now reset options, environment variables, the RNG kind, and the random seed afterward to what it was before being called.

  • Added package vignettes.

Bug Fixes

  • Tests on resolve() would use deprecated argument value.

Version 0.2.1 [2020-03-19]

CRAN Re-submission Requests

  • Update the package description to use single quotes.

  • Add example to check().

Version 0.2.0 [2020-03-13]

  • First version released on CRAN.

Version 0.1.1 [2020-01-06]

Bug Fixes

  • Assert that resolved() will launch lazy futures.

Version 0.1.0 [2020-01-03]

New Features

  • In non-interactive mode, check() will quit R with an exit code that reflects whether all tests passed (0) or not (1).

  • check() gained arguments so that it can be easily called from R too.

Version 0.0.0-9000 [2017-05-16]

  • Created package stub.