Fix the bridging header dependencies calculation for explicit module
build, especially for caching which needs an accurate list of deps for
compute cache key correctly.
Previously, the bridging header deps are computed from `ModuleGraph`
from the clang dependency scanner, which can be affected by already seen
modules. It causes the dependencies to be missing for bridging header if
the module is seen by main swift source module.
Now report only the directly module dependencies from bridging header,
then compute all the transitive dependencies before calculating all the
cache keys.
rdar://123156636
We previously blanket omitted `-Xcc -vfsoverlay` flags from Swift module dependencies' command-line recipes. This is incorrect as the Swift module must have an exact matching VFS overlay that its Clang dependencies use, in order to load said Clang dependnecies successfully and resolve their headers as expected and as was done during the scan.
Resolves rdar://122667530
Guard 'resetCache' and 'resetDiagnostics' as critical sections using 'DependencyScanningToolStateLock'. Otherwise there's a chance that one thread in the scanner is doing a reset on the diagnostic consumer, while some other thread is adding this diagnostic consumer to another scan instance which may also be populating said consumer at the time. Similarly, for resetting 'ScanningService' though it is much more unlikely to be reset while in-use by other scanning actions.
A swiftmodule can only be correctly ingested by a compiler
that has a matching state of using or not-using
NoncopyableGenerics.
The reason for this is fundamental: the absence of a Copyable
conformance in the swiftmodule indicates that a type is
noncopyable. Thus, if a compiler with NoncopyableGenerics
reads a swiftmodule that was not compiled with that feature,
it will think every type in that module is noncopyable.
Similarly, if a compiler with NoncopyableGenerics produces a
swiftmodule, there will be Copyable requirements on each
generic parameter that the compiler without the feature will
become confused about.
The solution here is to trigger a module mismatch, so that
the compiler re-generates the swiftmodule file using the
swiftinterface, which has been kept compatible with the compiler
regardless of whether the feature is enabled.
Previously, canImport lookup is not completely working with explicit
module due to two issues:
* For clang modules, canImport check still do a full modulemap lookup
which is repeated work from scanner. For caching builds, this lookup
cannot be performed because all modulemap and search path are dropped
after scanning.
* For swift module, if the canImport module was never actually imported
later, this canImport check will fail during the actual compilation,
causing different dependencies in the actual compilation.
To fix the problem, first unified the lookup method for clang and swift
module, which will only lookup the module dependencies reported by
scanner to determine if `canImport` succeed or not. Secondly, add all
the successful `canImport` check modules into the dependency of the
current module so this information can be used during actual
compilation.
Note the behavior change here is that if a module is only checked in
`canImport` but never imported still needs to be built. Comparing to
implicit module build, this can bring in additional clang modules if
they are only check inside `canImport` but should not increase work for
swift modules (where binary module needs to be on disk anyway) or the
most common usecase for `canImport` which is to check the same module
before importing.
rdar://121082031
Avoid path encoding difference (for example, real_path vs. path from
symlink) by eliminating the path from cache key. Cache key is now
encoded with the index of the input file from all the input files from
the command-line, reguardless if those inputs will produce output or
not. This is to ensure stable ordering even the batching is different.
Add a new cache computation API that is preferred for using input index
directly. Old API for cache key is deprecated but still updated to
fallback to real_path comparsion if needed.
As a result of swift scan API change, rename the feature in JSON file to
avoid version confusion between swift-driver and libSwiftScan.
rdar://119387650
Instead, only add the overlay itself, and let it refer to its own dependencies, which will still get recorded in the overall output.
Resolves rdar://117010118
The code, previously, only properly handled such dependencies being a distinct category for Swift source and Swift textual dependency infos. Swift binary module dependencies must handle this similarly and this change adds the missing support for them. Recent refactor of the scanner also means that now Swift binary dependencies with Swift overlay dependencies may crash the scanner, and this change resolves this as well.
Resolves rdar://117088840
Update swift cache key computation mechanism from one cache key per
output, to one cache key per primary input file (for all outputs that
associated with that input).
The new schema allows fewer cache lookups while still preserving most of
the flexibility for batch mode and incremental mode.
It is possible that import resolution failed because we are attempting to resolve a module which can only be brought in via a modulemap of a different Clang module dependency which is not otherwise on the current search paths. For example, suppose we are scanning a '.swiftinterface' for module 'Foo', which contains:
'''
@_exported import Foo
import Bar
...
Where 'Foo' is the underlying Framework clang module whose '.modulemap' defines an auxiliary module 'Bar'. Because 'Foo' is a framework, its modulemap is under '<some_framework_search_path>/Foo.framework/Modules/module.modulemap'. Which means that lookup of `Bar` alone from Swift will not be able to locate the module in it. However, the lookup of Foo will itself bring in the auxiliary module becuase the Clang scanner instance scanning for clang module Foo will be able to find it in the corresponding framework module's modulemap and register it as a dependency which means it will be registered with the scanner's cache in the step above. To handle such cases, we first add all successfully-resolved modules and (for Clang modules) their transitive dependencies to the cache, and then attempt to re-query imports for which resolution originally failed from the cache. If this fails, then the scanner genuinely failed to resolve this dependency.
Allow DependencyScanner to canonicalize path using a prefix map. When
option `-scanner-prefix-map` option is used, dependency scanner will
remap all the input paths in following:
* all the paths in the CAS file system or clang include tree
* all the paths related to input on the command-line returned by scanner
This allows all the input paths to be canonicalized so cache key can be
computed reguardless of the exact on disk path.
The sourceFile field is not remapped so build system can track the exact
file as on the local file system.
'ModuleDependencyScanner' maintains a Thread Pool along with a pool of workers
which are capable of executing a filesystem lookup of a named module dependency.
When resolving imports of a given Swift module, each import's resolution
operation can be issued asunchronously.
From being a scattered collection of 'static' methods in ScanDependencies.cpp
and member methods of ASTContext. This makes 'ScanDependencies.cpp' much easier
to read, and abstracts the actual scanning logic away to a place with common
state which will make it easier to reason about in the future.
In case import resolution order somehow sometimes matters, it's prudent to process/resolve/locate implicitly-imported modules first.
Resolves rdar://113917657
dependencies
It is valuable for clients to be able to distinguish which dependencies of a
Swift module originated from 'import' statements, and which ones are implicit
dependency Swift overlays of imported Clang modules.
Instead of the code querying the compiler's built-in Clang instance, refactor the
dependency scanner to explicitly keep track of module output path. It is still
set according to '-module-cache-path' as it has been prior to this change, but
now the scanner can use a different module cache for scanning PCMs, as specified
with '-clang-scanner-module-cache-path', without affecting module output path.
Resolves rdar://113222853
The code of `ScanDependencies.cpp` was creating invalid JSON since #66031
because in the case of having `extraPcmArgs` and `swiftOverlayDependencies`,
but not `bridgingHeader`, a comma will not be added at the end of
`extraPcmArgs`, creating an invalid JSON file. Additionally that same PR
added a trailing comma at the end of the `swiftOverlayDependencies`, which
valid JSON does not allow, but that bug was removed in #66366.
Both problems are, however, present in the 5.9 branch, because #66936
included #66031, but not #66366.
Besides fixing the problem in `ScanDependencies.cpp` I modified every test
that uses `--scan-dependencies` to pass the produced JSON through
Python's `json.tool` in order to validate proper JSON is produced. In
most cases I was able to pipe the output of the tool into `FileCheck`,
but in some cases the validation is done by itself because the checks
depend on the exact format generated by `--scan-dependencies`. In
a couple of tests I added a call to `FileCheck` that seemed to be
missing.
Without these changes, two tests seems to be generating invalid JSON in
my machine:
- `ScanDependencies/local_cache_consistency.swift` (which outputs `Expecting ',' delimiter: line 525 column 11 (char 22799)`)
- `ScanDependencies/placholder_overlay_deps.swift`
Add a flag `finalized` to indicate that a module entry in the dependency
cache is finalized and no longer needs to be updated. This prevents the
command-line flags from dependency inputs get added multiple times on
re-scan with the same service.
While during normal compilation, adding the same command-line flags
multiple times are fine, it is bad for caching builds as a new
compilation cache key needs to be computed every time.
Make sure the sources for bridging header is added as part of the CAS
filesystem for the main module. Even bridging header should be compiled
into PCH, PCHs are not standalone that it can be used without source
file.
Unlike `swift-frontend -scan-dependencies` option, when dependency
scanner is used as a library by swift driver, the SwiftScanningService
is shared for multiple driver invocations. It can't keep states (like
common file dependencies) that can change from one invocation to
another.
Instead, the clang/SDK file dependencies are computed from each driver
invocations to avoid out-of-date information when scanning service is
reused.
The test case for a shared Service will be added to swift-driver repo
since there is no tool to test it within swift compiler.
Reformatting everything now that we have `llvm` namespaces. I've
separated this from the main commit to help manage merge-conflicts and
for making it a bit easier to read the mega-patch.
This is phase-1 of switching from llvm::Optional to std::optional in the
next rebranch. llvm::Optional was removed from upstream LLVM, so we need
to migrate off rather soon. On Darwin, std::optional, and llvm::Optional
have the same layout, so we don't need to be as concerned about ABI
beyond the name mangling. `llvm::Optional` is only returned from one
function in
```
getStandardTypeSubst(StringRef TypeName,
bool allowConcurrencyManglings);
```
It's the return value, so it should not impact the mangling of the
function, and the layout is the same as `std::optional`, so it should be
mostly okay. This function doesn't appear to have users, and the ABI was
already broken 2 years ago for concurrency and no one seemed to notice
so this should be "okay".
I'm doing the migration incrementally so that folks working on main can
cherry-pick back to the release/5.9 branch. Once 5.9 is done and locked
away, then we can go through and finish the replacement. Since `None`
and `Optional` show up in contexts where they are not `llvm::None` and
`llvm::Optional`, I'm preparing the work now by going through and
removing the namespace unwrapping and making the `llvm` namespace
explicit. This should make it fairly mechanical to go through and
replace llvm::Optional with std::optional, and llvm::None with
std::nullopt. It's also a change that can be brought onto the
release/5.9 with minimal impact. This should be an NFC change.
Rename `-enable-cas` to `-compile-cache-job` to align with clang option
names and promote that to a new driver only flag.
Few other additions to driver flag for caching behaviors:
* `-compile-cache-remarks`: now cache hit/miss remarks are guarded behind
this flag
* `-compile-cache-skip`: skip replaying from the cache. Useful as a
debugging tool to do the compilation using CAS inputs even the output
is a hit from the cache.
