Summary:
The pest codegen has some non-determinism (HashMap) that breaks buck build.
According to jsgf:
This basically indicates some kind of non-determinism in the build. They're a pain, so I'd been hoping that we'd got them all.
Yeah, pest is generating non-deterministic output, which will screw things up badly. The problem is:
https://github.com/pest-parser/pest/blob/master/generator/src/generator.rs#L92-L93
fn generate_builtin_rules() -> HashMap<&'static str, TokenStream> {
let mut builtins = HashMap::new();
is putting builtins into a `HashMap`, then:
https://github.com/pest-parser/pest/blob/master/generator/src/generator.rs#L46
rules.extend(defaults.into_iter().map(|name| builtins[name].clone()));
emitting them in hashmap order. It needs to use a `BTreeMap` to make sure they're in a consistent order.
(I didn't check whether there are other instances of this.)
Reviewed By: jsgf
Differential Revision: D17063573
fbshipit-source-id: c03adc3c6d50bd09ffbd44ca8dc7bc51d6cad28d
Summary:
The workspace was added by D8741175 mainly to make build artifacts share a same
"target" directory. D14606468 made `setup.py` write a `.cargo/config` that
specifies a sharable "target" directory. Remove the workspace since the crates
already share a "target" directory.
This makes it a bit easier to add new crates.
Reviewed By: singhsrb
Differential Revision: D17053934
fbshipit-source-id: d34781c796356b725ddce3453c1951a4d4133807
Summary:
LTO does not buy us much - libbindings.so is 13MB disabled; 11MB enabled.
But LTO increases build time significantly:
- 24 seconds lto=false
- 36 seconds lto=thin
Therefore just disable LTO for faster builds.
Reviewed By: xavierd
Differential Revision: D17053932
fbshipit-source-id: efb437dbd2a26756bcc3f785901d06b286724835
Summary:
`ld.gold` is much faster than `ld.bfd`. That reduces `bindings` mtime-only change
debug build time from 10 seconds to 5 seconds.
Reviewed By: xavierd
Differential Revision: D17053933
fbshipit-source-id: 6d55a09723dd4d78d8a9db8c0721a18eabb6d435
Summary: This is already called in makeunionstores, no need to call it a second time.
Reviewed By: quark-zju
Differential Revision: D16794078
fbshipit-source-id: ff4c51d64748dfe4160a26767af11f7ebeb82618
Summary:
The Rust Manifests prevent having both a file and a directory with the same
name. This is a problem for the localrepo commit logic which will first add
the new entry, be that file or directory, and then remove the old one.
Reviewed By: quark-zju
Differential Revision: D16775350
fbshipit-source-id: 4515e42f9d2b1b2bc4861c16a39d291225cddc2a
Summary:
The insert code would be unclear in what kind of issue it ran into when
inserting files. Sometimes the file we want to insert is a directory and
other times it want to traverse a directory. This change makes those
situations clear along with some other corner case behaviors.
Reviewed By: quark-zju
Differential Revision: D16775354
fbshipit-source-id: 50ab6bc52b70cc5cef013d11050eb3cdf5b160a5
Summary:
Updating the manifest implementation for remove with the intended API.
When I originally implemented remove I wasn't confident what was the
best way to implement remove. As I've gained more experience, I feel
confident that doing two iteration over the tree is a good approach
for this method. The first iteration should validate that the file
exists then the second iteration will actually traverse down updating
the nodes to mutable ephemerals.
Reviewed By: quark-zju
Differential Revision: D16775353
fbshipit-source-id: 8ebee9ca347efcb694a6d27c1eeae2c149643766
Summary:
`get_link` started as test function that broaden in scope but did not have
it's behavior updated as it started to be used more broadly.
No reason to error out when we request a path that has parent files in the
manifest.
Reviewed By: quark-zju
Differential Revision: D16775356
fbshipit-source-id: a320926100378f16d723ca204746906e79c7752e
Summary:
There is no custom implementation for manfiest.dirs(). Generally speaking
the custom implementation is a good thing but doing the migration in the
current python codebase doesn't seem worth while at the moment.
Reviewed By: quark-zju
Differential Revision: D16775351
fbshipit-source-id: c428860d21088a50a0f754dc20d6ee224d2eae32
Summary:
The C++ constructs a manifestdict for matches. The assumption is that the
method is expected to return a manfiestdict and other code breaks if that
is broken. We will have to go and fix this later.
Reviewed By: quark-zju
Differential Revision: D16775352
fbshipit-source-id: 300945ccdd6448fc6ffae16a954aac93d6f526be
Summary:
We've seen problems where we send requests to the server for a manifest
node and claim that we have that same node as the base node. This results in the
server sending no data back. This is obviously a bug in the client, since if
we're trying to fetch the node it means we don't have it and shouldn't use it as
a base. Our current best guess is that the node is in the index but the actual
data is corrupt, so getmissing() claims it exists, but reads actually fail.
Until we figure out the root cause, let's prevent findrecenttree from returning
the node as a base to itself.
Reviewed By: sfilipco
Differential Revision: D16996425
fbshipit-source-id: 04b0cfbe1c432749f2aeb63f98c7b59ab50facc0
Summary:
Add blackbox logging of syncing to or from the commit cloud workspace,
including which changes are being submitted.
Also log when obsmarker fixup happens in blackbox.
Reviewed By: quark-zju
Differential Revision: D16961281
fbshipit-source-id: 0d0f675d77ab3446198703b31eea940dae3bdd85
Summary:
Add a debug log for when a repository is tracking visibility through visible
heads. It's not enough to just check if `visibility.enabled` is true, as the
repo needs to be migrated to visibility tracking, which might not have happened
yet.
Reviewed By: quark-zju
Differential Revision: D16961282
fbshipit-source-id: 8dec04dcc07e72e09f61399f9be04bdee339e262
Summary: This makes `config|debugconfig` as a debug command. Previously it was not.
Reviewed By: singhsrb
Differential Revision: D16866460
fbshipit-source-id: f1fcf0777d2fe6c1426f557ae1999710dd722109
Summary:
It is unclear where we generate basemfnodes that get to be equal to mfnodes
in our requests to the server. This logging allows us to track the codepath
that is problematic.
Reviewed By: quark-zju
Differential Revision: D16969888
fbshipit-source-id: b2d0aa7b573180664ad5cf8d6909451f089ac4a5
Summary:
Similar to the previous diff, opt out the revnum detection for the `repo.revs`
call used internally by directaccess to calculate revs.
Change smartlog to disable the revnum detection for its internal calculation.
Carefully choose the scope so `smartlog(1+2)` will still be warned.
Reviewed By: singhsrb
Differential Revision: D16954996
fbshipit-source-id: 2612055a8a9bb2fc9bdd1946c4bc565f967b22ca
Summary:
If `smartlog()` is the user input, the revnum detection is enabled for the
whole scope of smartlog revset implemenation. The use of rev numbers triggers
the warning. But the user didn't use any rev number.
Change smartlog to disable the revnum detection for its internal calculation.
Carefully choose the scope so `smartlog(1+2)` will still be warned.
Reviewed By: singhsrb
Differential Revision: D16954997
fbshipit-source-id: 1a3d32c1c2bcba08bfac908623b6416cf1cd63a8
Summary:
Instead of `debugsnapshot` and `debugcheckoutsnapshot` we will use
`snapshot create|checkout`.
Reviewed By: markbt
Differential Revision: D16961037
fbshipit-source-id: 142d072672376c2ce6a248a4dceffdd94d56a4df
Summary:
Aliases with `:doc` in name are not real commands. Do not treat them as
commands.
The upstream patch https://phab.mercurial-scm.org/D5087 added other metadata
including `:help` and `:category`. We might end up having some in the future
so I blacklisted names with `:` in them, not just `:doc`.
Reviewed By: chadaustin
Differential Revision: D16955596
fbshipit-source-id: b6f3e1129b632e0ba43c800e7e6fdd2b52d3b40c
Summary:
The type of `args` has changed. Tests did not capture the issue, since the code
block was not compiled using the default setup.
Reviewed By: sfilipco
Differential Revision: D16953736
fbshipit-source-id: a678f1354511f2d2ceb5faa7a4de0c0c6b03e84d
Summary:
`fs::canonicalize` has unwanted side effect - resolving symlinks.
Use `util::path::absolute` instead. This resolves a side effect where
Note on Linux, `chdir` resolves symlinks and `getcwd` (aka.
`std::env::current_dir`) will return the resolved path. On Windows
`std::env::current_dir` does not return the resolved path, at least
for the mapped drive created via the `subst` command.
Differential Revision: D16952484
fbshipit-source-id: 6969a1844020ff6b82de46416d8950ec40394159
Summary:
On Windows, some people use the "map drive" feature to map a long path (ex.
`C:\long\path\to\repo`) to a short path (ex. `Z:\`) so their tooling can
handle some long paths.
In that case, resolving symlinks by `hg root` is undesirable.
Unfortunately, the Rust stdlib does not have a Python `os.path.abspath`
equivalent. There were some attempts (ex. https://github.com/rust-lang/rust/pull/47363)
but the corner cases (ex. symlinks) have made the problem much more
complicated.
There are some 3rd-party crates. But they are not a good fit:
- https://github.com/danreeves/path-clean/ (last commit fb84930) follows the golang plan9 idea. It does not have proper support for Windows paths.
- https://github.com/vitiral/path_abs/ (latest commit 8370838) reinvents many path-related types, which is an overkill for this usecase.
This diff implements the feature "reasonably" for both Windows and Linux, with
nasty corner cases (symlink) ignored.
Differential Revision: D16952485
fbshipit-source-id: ba91f4975c2e018362e2530119765a380f103e19
Summary:
It does not seem to get used anywhere. Its usage of `scmutil.revrange` can
potentailly trigger a false positive "legacy revision number used" warning.
Reviewed By: chadaustin
Differential Revision: D16910635
fbshipit-source-id: 0b84d4b6a85678c581ccf1754bfbefbc97d8c055
Summary: The configparser crate is a better place for logic to load system and user config.
Reviewed By: sfilipco
Differential Revision: D16796396
fbshipit-source-id: c1ae5f85aa9839d89f5026279d699135dc1b442b
Summary:
Move `repo.shared_path` handling to repo initialization time and store it in
the repo structure.
This makes `repo.shared_path()` cheap if it gets used frequently.
Reviewed By: sfilipco
Differential Revision: D16796401
fbshipit-source-id: e19f3381cc87b55500ea1d27fd918ccb16a71972
Summary:
Some functions in `dispatch.rs` are about the "repo". Move them to a better
place - `repo.rs`.
The repo and config logic is coupled. A new enum `OptionalRepo` was added, to
make code easier to write - `Some(repo)` means the repo exists, and `repo` owns
the config. `None(config)` means the repo is missing, but the config is still
available.
Reviewed By: sfilipco
Differential Revision: D16796403
fbshipit-source-id: 2f4017a52296b629e990f85437b2cfdd7263b9e6
Summary:
"infer" means "try to get the repo path from command line arguments like a full path".
The enum variant is really "repo is optional". Rename to clarify.
Reviewed By: sfilipco
Differential Revision: D16796399
fbshipit-source-id: 505d2a406a83e0006200ece63d360b119548d2dd
Summary:
Change error type in clidispatch from `DispatchError` to `failure::Error`.
Pros:
- `failure` will attach a backtrace for free. (otherwise, backtrace handling is
manual)
- Wrapping other errors (ex. `io::Error`, `cliparser::Error`) is optional.
(otherwise, wrapping other errors is mandatory, and needs to be careful to
not lose information)
Cons:
- No longer able to enumerate *all* possible error types. (but can still
downcast to specific errors)
This seems to be a good tradeoff especially because of the backtrace handling - I
ran into a few issues where the location where the error happened really helped
debugging.
Since we can no longer enumerate all possible error types, the enum was changed
to individual structs to make the code shorter (ex. the struct can be downcasted
directly, instead of down-casting to the enum, then matching its variant).
The `HighLevelError` handling was simplified and moved to `hgmain`.
The new code path falls back to Python less aggressively, therefore some behaviors
were tweaked (ex. `-R` takes a bundle path).
Reviewed By: sfilipco
Differential Revision: D16796400
fbshipit-source-id: 1b17eb8d62503644b118c6b799778182bf222538
Summary:
This is more conistent with Mercurial style. And make them usable directly in
Rust code.
Reviewed By: sfilipco
Differential Revision: D16796397
fbshipit-source-id: 9016ea2b09fdf96b2b54138f5c8405caf96390f7
Summary:
This makes the "content" of the error stable. It is used in a later diff where
AmbiguousCommand gets handled by Rust directly instead of falling back to
Python dispatch.py.
Reviewed By: sfilipco
Differential Revision: D16796404
fbshipit-source-id: c439db14ec83c76c4762d3c627bfce1ea44bccf4
Summary:
In case there are many CommandDefinitions, we don't need all their flag
definitions until we decided to execute one of the commands.
Reviewed By: sfilipco
Differential Revision: D16796398
fbshipit-source-id: af205f59efd77fd7ff9eb4655d1f9167e2c350da
Summary: Convert global flags to `HgGlobalOpts` struct to make code shorter.
Reviewed By: sfilipco
Differential Revision: D16796407
fbshipit-source-id: b9d4c3dbec68c81908d439da4c353249347ca74a
Summary:
The tests are
- Mostly testing about configurations.
- Mostly depends on private functions.
ConfigSet already has a good test coverage, the tests do not provide much
value (aside from testing config override ordering, which is also covered
by hg tests). I'm going to change / remove the private functions. Remove the
tests to make changes easier.
Reviewed By: sfilipco
Differential Revision: D16796402
fbshipit-source-id: 56a8d55a0d1b0438bd2fcde5d3379d76f51dcd9d
Summary: This makes it possible to simplify the code reading global flags.
Reviewed By: sfilipco
Differential Revision: D16796405
fbshipit-source-id: eb604470d052ef84b748d1e60270cacd410fc5da
Summary:
The `Dispatcher` provides lots of features but its internal state only contains
the command table. Replace it with `CommandTable` and make the methods free
functions.
This makes function dependencies more cleaner, for example things like "locating
a repo", "getting the args" etc. won't require a `Dispatcher`.
A side effect of this change is the non-utf8 command line arguments are no longer
supported. This is already broken since our hg wrapper (accidentally) enforced
utf-8 command line. Therefore related tests are removed.
Reviewed By: sfilipco
Differential Revision: D16796395
fbshipit-source-id: a793ce7b8befe7caf62405c582fc932eb3daa099
Summary:
The dispatch logic couples with too much stuff - repo, parsing, command,
config. The command / registration part can be cleanly seprated.
This diff moves the core logic of CommandTable to command.rs.
Reviewed By: sfilipco
Differential Revision: D16796406
fbshipit-source-id: 5a33d6b6452537f07895bd9944122a3b3149d925
Summary:
At first, that aim was to create a good and consistent test plan of snapshots:
```
# Snapshot test plan:
# 1) Empty snapshot (no changes);
# 2) Snapshot with an empty manifest (changes only in tracked files);
# 3) Snapshot with a manifest (merge state + mixed changes);
# 4) Same as 3 but test the --clean flag on creation;
# 5) Same as 3 but test the --force flag on restore.
```
Then I discovered some bugs while writing the tests.
1) In case of an empty manifest the `snapshotmanifestid` extra should be an empty string, not `None`.
This is because of the extra serialization rules.
2) The initial approach for doing the checkout did some unnecessary operations with dirstate,
which required it to rollback later.
Now the overhead is gone, and the strategy of checking out is as follows:
```
1. Regular hg update to the (1st) parent of the snapshot;
2. Apply all the changes via mergemod.update without pushing it to the dirstate;
3. Register these changes in the dirstate;
4. Tie the second snapshot parent to the working context, if it exists.
Reviewed By: markbt
Differential Revision: D16929417
fbshipit-source-id: 0319ac0e81f80956cdef5b2a696b642cc782f48d
