Biweekly update from the GHC DevX team at IOG.
Previous updates can be found here.
JavaScript backend
Template Haskell
Sylvain: !9779 adding TH support for the JS backend passes CI and is ready for reviews.
Ticket #23013 has been opened to keep track of an issue with the recompilation avoidance mechanism. Fixing the issue seems to require some invasive refactoring that is best left for a future merge request.
Luite: !10008 Implemented keeping track of subdirectories in GHC's temporary file manager. Ready for review. Fixes #22952.
Temporary subdirectories are used when linking Template Haskell with the JavaScript backend and also in some situations when linking with other backends. This would always result in files being left behind in GHC's temporary directory (and a warning at high enough verbosity settings) since these subdirectories were never removed. With this patch, GHC keeps track of all created subdirectories and removes them at the end of the session.
JavaScript RTS refactor
Josh: has merged the refactor of the RTS generation module to reduce redundant code.
In this refactor, debug logging was also completed to determine the correct numbers
to use as the cache sizes for generated JavaScript names - allowing us to vastly
reduce the size of these arrays for efficiency, and to remove panic cases in favour
of generating higher numbered names without caching.
Integer performance
Sylvain: !9825 has been updated after an helpful review by Matthew Claven.
JavaScript EDSL
Jeff: MR is open. In response to feedback from the team Jeff has added (and removed) several features that distinguish the eDSL from sunroof. These include: removing threading and continuations, adding named unique variables and a proper switch statement. These changes allow more of the RTS to be replaced in the eDSL and allow the RTS to be typed using Haskell's type system. For example, now the STG registers track the type of the values they hold.
CI
Sylvain: fixed CI script test that prevented performance results to be stored for the JS backend (see #22923 and !10026).
Compiler performance
More-strict break/span
Josh: opened a CLC (Core Libraries Committee) proposal to add stricter versions of
break and span to Data.List in addition to the existing lazy versions. The
proposal considers evidence that these versions are situationally more performant,
by comparing allocation statistics, generated STG, and microbenchmarks - as well as
making the argument for consistency with existing List functions that also have
strict versions.
See also:
Unboxed CodeBuffers
Josh: opened a CLC proposal to modify the implementation of CodeBuffers in base
to use unboxed tuples in the return type of encoding functions. This change presents
a significant allocation improvement, due to the difficulty GHC has with applying a
certain optimisation within data types.
See also:
Optimization Handbook
Jeff: Has been hard at work on the Optimization Handbook.
He has finished a
chapter on Lambda Lifting, significantly expanded the glossary, and added
documentation to the
sphinx-exec-directive
haskell extension that he finished last month. The optimization handbook is now
in review by IOG's IT team to migrate it to the Haskell Foundation website.
GHCi
Luite: Added a testcase and did some cleanups in the types of the C code in MR
!9957, and adjusted
the bytecode generator to not produce zero offset SLIDE instructions anymore.
This is now ready for review.