tests: demonstrate crash when committing subset of copies to changeset When writing copy metadata to the changeset and not committing all copies in the dirstate, we get a ProgrammingError. This commit adds two tests showing how to trigger this bug. Differential Revision: https://phab.mercurial-scm.org/D6749

bdiff-torture: fix pyflakes warning reporting undefined name 'inst' Looks like I got a latest version of pyflakes somehow or it's running on py3 and it spotted this. Differential Revision: https://phab.mercurial-scm.org/D6757

split: handle partial commit of renames when doing split or record (issue5723) When using split or record, using either interface (text or curses), selecting portions of the file to be committed/recorded did not work; the entire file was treated as having been selected. This was because the logic for handling partial application of the patches relies on knowing what files are "new with modifications" and it doesn't treat "rename destination" as "new". There was a complicating issue, however. We're relying on the patch header specifying the copy from/to information, which works as long as the 'copy from' file is there. In the case of renames, however, the 'rename from' file is *not* there, so we need to add it back. Differential Revision: https://phab.mercurial-scm.org/D6768

split: handle partial commit of copies when doing split or record When using split or record, using either interface (text or curses), selecting portions of the file to be committed/recorded did not work; the entire file was treated as having been selected. This appears to be because the logic for handling partial application of the patches relies on knowing what files are "new with modifications", and it doesn't treat "copy destination" as "new". Handling renames correctly is more difficult and will be done in a later patch. Differential Revision: https://phab.mercurial-scm.org/D6767

py3: use pycompat.sysargv[0] for instead of fsencode(sys.argv[0]) Yuya noted in a recent review that fsencode(sys.argv[0]) could be incorrect on Windows. Differential Revision: https://phab.mercurial-scm.org/D6782

httppeer: use context manager when reading temporary bundle to send Differential Revision: https://phab.mercurial-scm.org/D6784

httppeer: use context manager when writing temporary bundle to send Differential Revision: https://phab.mercurial-scm.org/D6783

rust-cpython: mark unsafe functions as such It wasn't trivial to fix leak_immutable() to be safe since we have to allow immutable operations (e.g. iter()) on the leaked reference. So let's mark it unsafe for now. Callers must take care of the returned object to guarantee the memory safety. I'll revisit this later. I think $leaked<T: 'static> could have a function that converts itself into $leaked<U: 'static> with a given FnOnce(&T) -> &U, where T is $inner_struct, and U is $iterator_type for example.

rust-cpython: pair leaked reference with its manager object Still leak_immutable() is unsafe since leak_handle must live longer than the leaked_ref.

rust-cpython: introduce restricted variant of RefCell This should catch invalid borrow_mut() calls. Still the ref-sharing interface is unsafe.