narrow: fix flaky behavior described in issue6150 This has been plaguing the CI for a good while, and it doesn't appear to have an easy fix proposed yet. The solution in this change is to always do an unambiguous (but expensive) lookup in case of comparison. This should always be correct, albeit suboptimal. Differential Revision: https://phab.mercurial-scm.org/D10034

patch: make diff --git to differentiate b/w file is empty or doesn't exists Before this patch, as we didn't differentiate the two cases of a file in a context: 1. File doesn't exists 2. File is empty which causes the blob id to be same for both the cases. Now we use `nullhex` for a file which doesn't exists in a context (aligning it with the git diff format) Changes in test file reflect the fixed behavior. Differential Revision: https://phab.mercurial-scm.org/D10001

tests: add a test to demonstrate a bug in `hg diff --git` (issue6486) Issue url: https://bz.mercurial-scm.org/show_bug.cgi?id=6486 This will be fixed in next patch. Differential Revision: https://phab.mercurial-scm.org/D10000

test-copies: reinstall initial identical (empty) files for chained copied This effectively back out changeset deeb215be337. Changeset deeb215be33 does not really include a justification for its change and make mes uncomfortable. I have been thinking about it and they are two options: - either having empty/full files does not make a difference, and deeb215be337 is a gratuitous changes. - either having empty/full files do make a difference and deeb215be33 silently change the test coverage. In such situation if we want the "not empty" case to be tested, we should add new cases to cover them In practice, we know that the "file content did not change, but merge still need to create a new filenode" case exists (for example if merging result in similar content but both parent of the file need to be recorded), and that such case are easy to miss/mess-up in the tests. Having all the file using the same (empty) content was done on purpose to increase the coverage of such corner case. As a result I am reinstalling the previous test situation. To increase the coverage of some case involving content-merge in test-copies-chain-merge.t, we will add a new, dedicated, cases later in this series, once various cleanup and test improvement have been set in place. This changeset starts with reinstalling the previous situation as (1) it is more fragile, so I am more confided getting it back in the initial situation, (2) I have specific test further down the line that are base on these one. The next changeset will slightly alter the test to use non-empty files for these tests (with identical content). It should help to make the initial intent "merge file with identical content" clearer. I am still using a two steps (backout, then change content) approach to facilitate careful validation of the output change. Doing so has a large impact on the output of the "copy info in changeset extra" variant added in 5e72827dae1e (2 changesets after deeb215be33). It seems to highlight various breakage when merge without content change are involved, this is a good example of why we want to explicitly test theses cases. Because the different -do- matters a lot. Fixing the "copy info in changeset extra" is not a priority here. Because (1) this changeset does not break anything, it only highlight that they were always broken. (2) the only people using "copy info in changeset extra" do not have merge. Differential Revision: https://phab.mercurial-scm.org/D9587

upgrade: speed up when we have only nodemap to downgrade Similar to what we do on upgrade, if we have only persistent-nodemap to downgrade we will just delete the nodemap files and update repository requirements instead of processing all the revlogs. After downgrade, we are left with unrequired docket and transaction files which seems fine but can work on deleting them if someone feels we should. Differential Revision: https://phab.mercurial-scm.org/D9992

upgrade: write nodemap for manifests too In 98e39f04d60e I assumed that writing nodemap for manifests was not desirable and stopped writing it during upgrade. However in recent discussion with Pierre-Yves, I learnt that that's not true. Differential Revision: https://phab.mercurial-scm.org/D9991

windows: fix parsing of version number to match format from D9955 Differential Revision: https://phab.mercurial-scm.org/D10061

build: make version from .hg_archival.txt consistent with that from .hg/ D9955 changed the version format to replace "-" by "." and to add "hg" before the number representing the distance from the latest tag. However, it missed the "hg" string and added an extra "." to the version string we produce when there's a `.hg_archival.txt`. This patch makes it consistent. Differential Revision: https://phab.mercurial-scm.org/D10060

helptext: fix a recent typo Differential Revision: https://phab.mercurial-scm.org/D10033

re2: byteify some regex used to get check re2 availability Changeset 687b865b95ad failed to properly byteify this because it was a raw string. This went undetected so far because re2 does not seems to be widely tested. Differential Revision: https://phab.mercurial-scm.org/D10064