hook: ignore EPIPE when flushing stdout/stderr This fixes the bug described in the parent commit. test-transaction-rollback-on-sigpipe.t is updated to show the new behavior. Differential Revision: https://phab.mercurial-scm.org/D9152

test: add test-transaction-rollback-on-sigpipe.t demonstrating py3 regression When an hg push is interrupted with C-c, the remote [hg serve] command receives SIGPIPE. If a pretxnchangegroup hook fails, the remote hg then tries to rollback the transaction. It begins by printing "transaction abort!\n". This returns EPIPE, but ui.py ignores that error. In python3 (but not python2), this "transaction abort!\n" message stays in a buffer, so future flushes of stderr will try to print the message again, and so those flushes will also hit EPIPE. This test demonstrates such a case where this EPIPE causes the transaction rollback to fail, leaving behind an abandoned transaction. Differential Revision: https://phab.mercurial-scm.org/D9151

upgrade: allow sidedata upgrade to modify revision flag In the process, we fix the lack of HAS_COPIES_INFO flag on upgrade, and test the results. Differential Revision: https://phab.mercurial-scm.org/D9199

sidedata: return enough data to set the proper flag in the future If the revision has information relevant to copy tracing, we need to set a dedicated flag in revlog. Currently the upgrade process is failing to do so. Before we teach the upgrade process about flags, we make the information available where we will needs it. Differential Revision: https://phab.mercurial-scm.org/D9198

test: move upgrade run and check earlier in test-copies-chain-merge We are about to introduce an explicit case for copy tracing after a upgrade. So I am moving the code around beforehand for clarity. Differential Revision: https://phab.mercurial-scm.org/D9196

copies: split creation of the graph and actual checking again The re-install the old split. It will be necessary to test that the upgrade process produced a functionally identical result. It will be useful to detect case where the metadata we look at identical, but some other items we did not checked are missing. (spoiler: we will find some bug) Differential Revision: https://phab.mercurial-scm.org/D9195

copy: clarify in help text that `hg co --forget` takes a *destination* file We had a user who tried to pass a source file. The command then fails with `<file>: not unmarking as copy - file is not marked as copied`, so at least it's not just silent, but let's be a little clearer in the documentation. Differential Revision: https://phab.mercurial-scm.org/D9214

test: check server error output in `test-pull-bundle.t` This is useful to debug test failure when they happens. Differential Revision: https://phab.mercurial-scm.org/D9213

unbundle: free temporary objects after use This reduces peak RSS for larger unbundle operations by ~30 Bytes per changeset on AMD64. This can't be a direct delete for Python 2.7, so reset the object instead and leave a comment. The efilesset object can't be deleted as it is referenced by the local onchangelog function and Python 2.7 rejects a delete on the existance of a nested scope. Differential Revision: https://phab.mercurial-scm.org/D9153

changing-files: add a shorthand property to check for copy relevant info We are going to reuse this change in more place, so we factor it out first. Differential Revision: https://phab.mercurial-scm.org/D9197