workingctx: build _manifest on filenode() or flags() request I'm not sure if this is the best workaround, but this fixes the following exception: AttributeError: 'workingctx' object has no attribute '_manifestdelta' The short hash '303030303030' seen in the test is node.modifiednodeid. Differential Revision: https://phab.mercurial-scm.org/D2940

tests: enter full hex hash in plain text in bundle part We were looking it up be prefix by repo.__getitem__, which I'm about to drop support for. It's easiest to just include the full hash in plain text. Differential Revision: https://phab.mercurial-scm.org/D3168

tests: call rawsize() directly rawsize() is not reimplemented outside of revlog. I'm not sure why this code was insisting it call a specific implementation. Changing it to call rawsize() on the repo.file(f) result seems to work just fine. Differential Revision: https://phab.mercurial-scm.org/D3153

upgrade: sniff for filelog type The upgrade code should never encounter a vanilla revlog instance: only changelog, manifestrevlog, and filelog should be seen. The previous code assumed !changelog & !manifestrevlog meant file data. So this change feels pretty safe. If nothing else, it will help tease out typing issues. Differential Revision: https://phab.mercurial-scm.org/D3152

revlog: move censor logic into main revlog class Previously, the revlog class implemented dummy methods for various censor-related functionality. Revision censoring was (and will continue to be) only possible on filelog instances. So filelog implemented these methods to perform something reasonable. A problem with implementing censoring on filelog is that it assumes filelog is a revlog. Upcoming work to formalize the filelog interface will make this not true. Furthermore, the censoring logic is security-sensitive. I think action-at-a-distance with custom implementation of core revlog APIs in derived classes is a bit dangerous. I think at a minimum the censor logic should live in revlog.py. I was tempted to created a "censored revlog" class that basically pulled these methods out of filelog. But, I wasn't a huge fan of overriding core methods in child classes. A reason to do that would be performance. However, the censoring code only comes into play when: * hash verification fails * delta generation * applying deltas from changegroups The new code is conditional on an instance attribute. So the overhead for running the censored code when the revlog isn't censorable is an attribute lookup. All of these operations are at least a magnitude slower than a Python attribute lookup. So there shouldn't be a performance concern. Differential Revision: https://phab.mercurial-scm.org/D3151

revlog: move parsemeta() and packmeta() from filelog (API) filelog.parsemeta() and filelog.packmeta() are used to decode and encode metadata for file copies and censor. An upcoming commit will move the core logic for censoring revlogs into revlog.py. This would create a cycle between revlog.py and filelog.py. So we move these metadata functions to revlog.py. .. api:: filelog.parsemeta() and filelog.packmeta() have been moved to the revlog module. Differential Revision: https://phab.mercurial-scm.org/D3150

filelog: declare that filelog implements a storage interface Now that we have a declared interface, let's declare that filelog implements it. Tests have been added that confirm the object conforms to the interface. The existing interface checks verify there are no extra public attributes outside the declared interface. filelog has several extra attributes. So we added a mechanism to suppress this check. The goal is to modify the filelog class so we can drop this check. Differential Revision: https://phab.mercurial-scm.org/D3149

repository: define existing interface for file storage Now that we have mostly successfully implemented an alternate storage backend for files data, let's start to define the interface for it! This commit takes the mostly-working interface as defined by the simple store repo and codifies it as the file storage interface. The interface has been split into its logical components: * index metadata * fulltext data * mutation * everything else I don't consider the existing interface to be great. But it will help to have it more formally defined so we can start chipping away at refactoring it. Differential Revision: https://phab.mercurial-scm.org/D3148

tests: run some largefiles and lfs tests with simple store Now that the simple store handles flags properly, a handful of the largefiles and lfs tests pass! Differential Revision: https://phab.mercurial-scm.org/D3147

commands: don't violate storage abstractions in `manifest --all` Previously, we asked the store to emit its data files. For modern repos, this would use fncache to resolve the set of files then would stat() each file. For my copy of the mozilla-unified repository, this took 3.3-10s depending on the state of my filesystem cache to render 449,790 items. The previous behavior was a massive layering violation because it assumed tracked files would have specific filenames in specific directories. Alternate storage backends would violate this assumption. The new behavior scans the changelog entries for the set of files changed by each commit. It aggregates them into a set and then sorts and prints the result. This reliably takes ~16.3s on my machine. ~80% of the time is spent in zlib decompression. The performance regression is unfortunate. If we want to claw it back, we can create a proper storage API to query for the set of tracked files. I'm not opposed to doing that. But I'm in no hurry because I suspect ~0 people care about the performance of `hg manifest --all`. .. perf:: `hg manifest --all` is likely slower due to changing its implementation to respect storage interface boundaries. If you are impacted by this regression in a meaningful way, please make noise on the development mailing list and it can be dealt with. Differential Revision: https://phab.mercurial-scm.org/D3119