…y on all platforms.
… more build now
DataRowView indexer should not break on "completed" added rows (any rows other than the IsNew one)
../../libgc/alloc.c: In function ‘GC_stopped_mark’: ../../libgc/alloc.c:501:9: error: implicit declaration of function ‘GC_cond_register_dynamic_libraries’ [-Werror=implicit-function-declaration] GC_cond_register_dynamic_libraries();
This accounting mode logs the "cost" of each bridge-object during bridge processing. This should give a very good first clue on how expensive each object is.
…ation. My previous changes modified the way how we're handing the request body when resending a request due to a 30x redirection. Authentication is a "special" form of redirection because we sometimes need to resend a request with additional authentication headers. Unfortunately, I forgot this, causing a regression which is now fixed. This has been reported as bugs #19119 and #19068.
… unsupported IL opcode
…83. Thanks to Bill Seurer for spotting it.
Travis simply doesn't give us anywhere near enough execution time to run runtime tests + BCL tests.
This removes a lot of duplicated logic and special-casing for Wrench. There are now only 3 important top-level targets: * `check`: Run all stable (~fast) tests. * `check-full`: Run all stable, unstable, and slow tests. * `check-parallel`: Compile tests in parallel, then run `check-full`. By default, `no-dump` is passed to the test runner. To pass `dump-output` and get `stdout`/`stderr` dumped to console when a failure occurs, set `V` as in: $ make check V=1 Or: $ make check-full V=1 If you're running tests in a CI environment, it may be desirable to disable tests that are known to be unstable. To do so, run: $ make check CI=1 Or: $ make check-full CI=1
…ry time there is a public api break. This allows the llvm backend to compile with different versions of LLVM.
When we want to iterate without sweeping, then on blocks that have not been swept, we can only iterate over objects that are marked.
We end up with faux arrays when we’re allocating new objects, but the nursery must be zeroed. We just zero the first few words when the TLAB is created to get rid of the array header. An alternative would be to erase the headers once the nursery collection is over, which is cleaner and would only zero what we know must be zeroed, i.e. it would catch errors in cases where we now overwrite non-zero garbage in the nursery. That’s an unlikely case, though, and we don’t use this debug feature often anyway.