- Nov 25, 2024
-
-
This patch makes test-bootstrap related ci jobs only depend on hadrian-ghc-in-ghci job to finish, consistent with other jobs in the full-build stage generated by gen_ci.hs. This allows the jobs to be spawned earlier and improve overall pipeline parallelism.
-
This patch makes some minor improvements re nix-in-docker logic in the ci configuration: - Update `nixos/nix` to the latest version - Apply $CPUS to `cores`/`max-jobs` to avoid oversubscribing while allowing a reasonable degree of parallelism - Remove redundant `--extra-experimental-features nix-command` in later `nix shell` invocations, it's already configured in `/etc/nix/nix.conf`
-
-
The default value for base-unit-id is stored in the settings file. At install time, this can be set by using the BASE_UNIT_ID environment variable. At runtime, the value can be set by `-base-unit-id` flag. For whether all this is a good idea, see #25382 Fixes #25382
-
-
As noted in #25509, the `-this-package-name` must be passed for each package to ensure that GHC can response references to the packages' exposed modules via package-qualified imports. Fix this. Closes #25509.
-
- Nov 23, 2024
-
-
-
-
-
-
Some testsuite tests require specific tools (e.g. `check-ppr` and `check-exact`) beyond those shipped in the binary distribution. Skip these tests. Fixes #13897.
-
With the removal of the duplicate backtrace, part of CLC proposal #285, the constructor `ErrorCallWithLocation` was removed from base. This commit re-introduces it with a deprecation.
-
This patch adds a missing Show constraint to a code example in the User Guide's type abstractions docs to fix issue #25422.
-
This patch removes the unused `GHC.Linker.Loader.loadExpr` function. It was moved from `GHC.Runtime.Linker.linkExpr` in `ghc-9.0` to `GHC.Linker.Loader.loadExpr` in `ghc-9.2`, and remain completely unused and untested ever since. There's also no third party user of this function to my best knowledge, so let's remove this. Anyone who wants to write their own GHC API function to load bytecode can consult the source code in older release branches.
-
Marks the following tests as broken on i386/Alpine: * T22033 due to #25497 * simd009, T25062_V16, T25169, T22187_run due to #25498
-
-
As requested by Mikolaj and started by Julian. Co-Authored-By:Julian Ospald <hasufell@hasufell.de>
-
For introduction of Alpine/i386 image. Thanks to Julian for the base image. Co-Authored-By:
-
- Nov 21, 2024
-
-
This commit fixes an undefined symbol error in RTS linker when attempting to compile home modules with -fhpc and -fbyte-code-and-object-code/-fprefer-byte-code, see #25510 for detailed description and analysis of the bug. Also adds T25510/T25510c regression tests to test make mode/oneshot mode of the bug.
-
Introduces `ReachabilityIndex`, an index constructed from a `GHC.Data.Graph.Directed` `Graph` that supports fast reachability queries (in $O(1)$). This abstract data structure is exposed from `GHC.Data.Graph.Directed.Reachability`. This index is constructed from the module graph nodes and cached in `ModuleGraph`, enabling efficient reachability queries on the module graph. Previously, we'd construct a Map of Set of ModuleGraph nodes which used a lot of memory (`O(n^2)` in the number of nodes) and cache that in the `ModuleGraph`. By using the reachability index we get rid of this space leak in the module graph -- even though the index is still quadratic in the number of modules, it is much, much more space efficient due to its representation using an IntMap of IntSet as opposed to the transitive closure we previously cached. In a memory profile of MultiLayerModules with 100x100 modules, memory usage improved from 6GB residency to 2.8GB, out of which roughly 1.8GB are caused by a second space leak related to ModuleGraph. On the same program, it brings compile time from 7.5s to 5.5s. Note how we simplify `checkHomeUnitsClosed` in terms of `isReachableMany` and by avoiding constructing a second graph with the full transitive closure -- it suffices to answer the reachability query on the full graph without collapsing the transitive closure completely into nodes. Unfortunately, solving this leak means we have to do a little bit more work since we can no longer cache the result of turning vertex indices into nodes. This results in a slight regression in MultiLayerModulesTH_Make, but results in large performance and memory wins when compiling large amounts of modules. ------------------------- Metric Decrease: mhu-perf Metric Increase: MultiLayerModulesTH_Make ------------------------- -
-
- Nov 20, 2024
-
-
This commit fixes the implementation of 'regUsageOfInstr' for vector operations that take an 'Operand' as the destination, by ensuring that when the destination is an address then the address should be *READ*, and not *WRITTEN*. Getting this wrong is a disaster, as it means the register allocator has incorrect information, which can lead to it discard stores to registers, segfaults ensuing. Fixes #25486
-
Arnaud Spiwack authored
Otherwise it's not scaled properly by the context, allowing unsound expressions. Fixes #25185.
-
Arnaud Spiwack authored
-
- Nov 19, 2024
-
-
See #25350 for floating-point min/max Co-authored-by:sheaf <sam.derbyshire@gmail.com>
-
With new mkHomeModLocation that takes an extra HscSource to add boot extensions if required.
-
Partially reverts commit fff55592 Amends add(Home)ModuleToFinder so that locations for boot files are not stored in the finder cache. Removes InstalledModule field from InstalledFound constructor since it's the same as the key that was searched for.
-
This is a convenience utility for use in GDB.
-
-
- Nov 18, 2024
-
-
We were reserving the wrong kind of adjustor context (the generic `AdjustorContext` used by other adjustor implementations, rather than the i386-specific `CCallContext`) to return the adjustor context while freeing, resulting in #25485. Fixes #25485.
-
-
The testsuites of array and stm had to be updated according to !13301. Updates submodule array and stm.
-
-
When running the program with `runhaskell`/`runghc` the backtrace should match the backtrace one would get by compiling and running the program. But currently, an exception thrown in a program interpreted with `runhaskell` will: * Not include the original exception backtrace at all * Include the backtrace from the internal GHCi/ghc rethrowing of the original exception This commit fixes this divergence by not annotating the ghc(i) backtrace (with NoBacktrace) and making sure that the backtrace of the original exception is serialized across the boundary and rethrown with the appropriate context. Fixes #25116 The !13301 MR (not this commit in particular) improves performance of MultiLayerModules. Unfortunately, T3294 regresses on aarch64-linux-deb12 by 1% allocations. Since this patch must be merged for 9.12 ASAP, we will not be able to investigate the slight regression on this platform in time. ------------------------- Metric Decrease: MultiLayerModulesRecomp MultiLayerModulesTH_OneShot Metric Increase: T3294 ------------------------- -
This was originally proposed and accepted in section "2.7 Capturing Backtraces on Exceptions" of the CLC proposal for exception backtraces. However, the implementation missed this re-export, which this commit now fixes. -
The former was unused except for one module where it was essentially re-defining displayExceptionContext. Moreover, this commit extends the fix from bfe600f5 to displayExceptionContext too, which was missing.
-
CLC proposal#285
-
Before the exception backtraces proposal was implemented, ErrorCall accumulated its own callstack via HasCallStack constraints, but ExceptionContext is now accumulated automatically. The original ErrorCall mechanism is now redundant and we get a duplicate CallStack Updates Cabal submodule to fix their usage of ErrorCallWithLocation to ErrorCall CLC proposal#285 Fixes #25283
-
This commit changes the Exception instance of SomeException to *simply* display the underlying exception in `displayException`. The augmented exception message that included the type and backtrace of the exception are now only printed on a call to `displayExceptionWithInfo`. At a surface level, existing programs should behave the same since the `uncaughtExceptionHandler`, which is responsible for printing out uncaught exceptions to the user, will use `displayExceptionWithInfo` by default. However, unlike the instance's `displayException` method, the `uncaughtExceptionHandler` can be overriden with `setUncaughtExceptionHandler`. This makes the extra information opt-in without fixing it the instance, which can be valuable if your program wants to display uncaught exceptions to users in a user-facing way (ie without backtraces). This is what was originally agreed for CLC#231 or CLC#261 with regard to the type of the exception information. The call stack also becoming part of the default handler rather than the Exception instance is an ammendment to CLC#164. Discussion of the ammendment is part of CLC#285.
-
This commit fixes the layout of the additional information included when displaying an exception, namely the type of the exception. It also fixes the default handler's heading message to work well together with the improved display message of SomeException. CLC proposal#285
-
