This reverts commit 4f059033bb. The change
is actually not NFC since previously, there is a cache in the
CompilerInvocation that prevents the same CAS from the same CASOptions
from being initialized multiple times, which was relied upon when
running inside sub invocation. When switching to a non-caching simple
CASOption types, it causes every single sub instance will create its own
CAS, and it can consume too many file descriptors and causing errors
during dependency scanning.
rdar://164903080
Previously, with the change to bridge Clang dependency scanning results on-demand, the scanner would execute Clang dependency scanning queries for each unresolved import, in parallel, and aggregate all of the results to post-process (including on-demand bridging) later. As a consequence of that change, all of the Clang scanner queries' results ('TranslationUnitDeps') got aggregated during a scan and had their lifetimes extended until a later point when they got processed and added to the scanner's cache.
This change refactors the Clang dependency scanner invocation to, upon query completion, accumulate only the 'ModuleDeps' nodes which have not been registered by a prior scan, discarding the rest of the 'TranslationUnitDeps' graph. The arrgegated 'ModuleDeps' objects are still bridged on-demand downstream.
This change further splits up the 'resolveAllClangModuleDependencies' method's functionality to improve readability and maintainability, into:
- 'gatherUnresolvedImports' method which collects all of collected Swift dependents' imports which did not get resolved to Swift dependencies
- 'performParallelClangModuleLookup' which actually executes the parallel queries and includes the new logic described above
- 'cacheComputedClangModuleLookupResults' method which takes the result of the parallel Clang scanner query and records in in the Swift scanner cache
- 'reQueryMissedModulesFromCache' method which covers the scenario where Clang scanner query returned no result because either the dependency can only be found transitively, or the query is for a dependency previously-queried.
This moves the functionality of 'bridgeClangModuleDependency' into a utility in the main scanner class because it relies on various objects whose lifetime is already tied to the scanner itself.
Previously, frequently-used methods like 'getAllDependencies' and 'getAllClangDependencies' had to aggregate (copy) multiple collections stored in a 'ModuleDependencyInfo' into a new result array to present to the client. These methods have been refactored to instead return an iterable joined view of the constituent collections.
This is required in order to always have computed the set of visible Clang modules for each Swift module in the graph. Otherwise when some Clang module gets cached as a transitive dependency from a query and is later looked up as a direct dependency, there will not be any computed visible modules set.
Previously Swift overlay lookup was performed for every directly and transitively-imported Clang module.
https://github.com/llvm/llvm-project/pull/147969 introduced the concept of "visible" Clang modules from a given named Clang dependency scanner query which closely maps to the set of modules for which Swift will attempt to load a Swift overlay. This change switches overlay querying to apply only to the set of such visible modules.
Resolves rdar://144797648
The note will point the user to where the "other" module with the same name is located and mention whether it is an SDK module. This is nice to have in various circumstances where developers attempt to define a module with the same name as a Swift module that already exists on their search paths, for example in the SDK.
Move per-query state out of ScanningService. There is still a check to
make sure the CASOptions are matching between queries because of the
requirement on clang scanner. Otherwise, the scanning service should
contain no per-query information anymore.
Resolves: https://github.com/swiftlang/swift/issues/82490
While this made sense in the distant past where the scanning service provided backing storage for the dependency cache, it no longer does so and now makes for awkard layering where clients get at the service via the cache. Now the cache is a simple data structure while all the clients that need access to the scanning service will get it explicitly.
- 'SwiftModuleScanner' will now be owned directly by the 'ModuleDependencyScanningWorker' and will contain all the necessary custom logic, instead of being instantiated by the module interface loader for each query
- Moves ownership over module output path and sdk module output path directly into the scanning worker, instead of the cache
This was used a long time ago for a design of a scanner which could rely on the client to specify that some modules *will be* present at a given location but are not yet during the scan. We have long ago determined that the scanner must have all modules available to it at the time of scan for soundness. This code has been stale for a couple of years and it is time to simplify things a bit by deleting it.
Adds an access control field for each imported module identified. When multiple imports of the same module are found, this keeps track of the most "open" access specifier.
On creation, 'ClangImporter' adds overlay modulemap files for non-modular platform libraries (e.g. glibc, libstdc++), which allows Swift code to import and use those libraries.
This change adds the same filesystem overlay to dependency scanning queries by applying them to the filesystem instantiated for each depndency scanning worker. Without these overlays EBM builds cannot discover and use non-modular system libraries on non-Darwin platforms.
Resolves rdar://151780437
In expectation, this should never happen. Such a situation means that within the same scanning action, Clang Dependency Scanner has produced two different variants of the same module. This is not supposed to happen, but we are currently hunting down the rare cases where it does, seemingly due to differences in Clang Scanner direct by-name queries and transitive header lookup queries.
Improve diagnostics message for swift caching build by trying to emit
the diagnostics early when there is more context to differentiate the
different kind of problems.
After the improvement, CAS Error should be more closer to when there is
functional problem with the CAS, rather than mixing in other kinds of
problem (like scanning dependency failures) when operating with a CAS.
rdar://145676736
With '-sdk-module-cache-path', Swift textual interfaces found in the SDK will be built into a separate SDK-specific module cache.
Clang modules are not yet affected by this change, pending addition of the required API.
The algorithm already performs pairwise checks on module dependencies brought into compilation per-source-file. Previously, the algorithm considered the entire sub-graph of a given source file. Actual source compiles do not consider the full transitive module dependency set for cross-import-overlay lookup, but rather only directly-imported modules in a given source file, and '@_exported import' Swift transitive dependencies.
This change adds tracking of whether a given import statement is 'exported' to the dependency scanner and then refines the cross-import overlay lookup logic to only consider transitive modules that are exported by directly-imported dependencies.
Add ability to automatically chaining the bridging headers discovered from all
dependencies module when doing swift caching build. This will eliminate all
implicit bridging header imports from the build and make the bridging header
importing behavior much more reliable, while keep the compatibility at maximum.
For example, if the current module A depends on module B and C, and both B and
C are binary modules that uses bridging header, when building module A,
dependency scanner will construct a new header that chains three bridging
headers together with the option to build a PCH from it. This will make all
importing errors more obvious while improving the performance.
Checking each module dependency info if it is up-to-date with respect to when the cache contents were serialized in a prior scan.
- Add a timestamp field to the serialization format for the dependency scanner cache
- Add a flag "-validate-prior-dependency-scan-cache" which, when combined with "-load-dependency-scan-cache" will have the scanner prune dependencies from the deserialized cache which have inputs that are newer than the prior scan itself
With the above in-place, the scan otherwise proceeds as-is, getting cache hits for entries still valid since the prior scan.
Instead, each scan's 'ModuleDependenciesCache' will hold all of the data corresponding to discovered module dependencies.
The initial design presumed the possibility of sharing a global scanning cache amongs different scanner invocations, possibly even different concurrent scanner invocations.
This change also deprecates two libSwiftScan entry-points: 'swiftscan_scanner_cache_load' and 'swiftscan_scanner_cache_serialize'. They never ended up getting used, and since this code has been largely stale, we are confident they have not otherwise had users, and they do not fit with this design.
A follow-up change will re-introduce moduele dependency cache serialization on a per-query basis and bring the binary format up-to-date.
This change refactors the top-level dependency scanning flow to follow the following procedure:
Scan():
1. From the source target under scan, query all imported module identifiers for a *Swift* module. Leave unresolved identifiers unresolved. Proceed transitively to build a *Swift* module dependency graph.
2. Take every unresolved import identifier in the graph from (1) and, assuming that it must be a Clang module, dispatch all of them to be queried in-parallel by the scanner's worker pool.
3. Resolve bridging header Clang module dpendencies
4. Resolve all Swift overlay dependencies, relying on all Clang modules collected in (2) and (3)
5. For the source target under scan, use all of the above discovered module dependencies to resolve all cross-import overlay dependencies
Use IncludeTreeFileList instead of full feature CASFS for swift
dependency filesystem. This allows smaller CAS based VFS that is smaller
and faster. This is enabled by the CAS enabled compilation does not
need to iterate file system.
rdar://136787368
Add function to handle all macro dependencies kinds in the scanner,
including taking care of the macro definitions in the module interface
for its client to use. The change involves:
* Encode the macro definition inside the binary module
* Resolve macro modules in the dependencies scanners, including those
declared inside the dependency modules.
* Propagate the macro defined from the direct dependencies to track
all the potentially available modules inside a module compilation.
When scanning swift modules and constructing their build commands, there
is no need to pass any external plugin search paths if there are no macro
dependencies for the module.
rdar://135221984
This makes sure that Swift respects `-Xcc -stdlib=libc++` flags.
Clang already has existing logic to discover the system-wide libc++ installation on Linux. We rely on that logic here.
Importing a Swift module that was built with a different C++ stdlib is not supported and emits an error.
The Cxx module can be imported when compiling with any C++ stdlib. The synthesized conformances, e.g. to CxxRandomAccessCollection also work. However, CxxStdlib currently cannot be imported when compiling with libc++, since on Linux it refers to symbols from libstdc++ which have different mangled names in libc++.
rdar://118357548 / https://github.com/swiftlang/swift/issues/69825
Rather than only protecting the insertion and non-const access to
`ContextSpecificCacheMap` in ScanningService, extend the mutex
protection to all accesses. Even a 'const' lookup in the cache map is
not thread safe because the `StringMap` could be in the process of being
rehashed.
rdar://127205953
