Table of Contents

Today, the Chapel developer community is pleased to announce the release of version 1.29.0 of Chapel! To obtain a copy, please refer to the Downloading Chapel page on the Chapel website.

Highlights of Chapel 1.29.0

Compilation Times

This version of Chapel includes a change to how the chpl compiler is built, causing it to use jemalloc by default on all platforms other than Mac and Cygwin. We’ve found this to result in a significant compile-time savings on average, as can be seen on October 10th in our nightly performance graph tracking average compilation times across the Chapel test suite.

Optimized Performance

In terms of the performance of Chapel programs themselves, version 1.29.0 includes improvements to the performance and scalability of creating distributed domains and arrays. In practice, these improvements have been shown to improve execution time of user applications that create and destroy arrays frequently, such as Arkouda.

Library Stabilization

A large number of the changes in Chapel 1.29.0 involve continued improvements to our standard library modules in terms of renamings, behavior improvements, and the like as part of our preparation for the forthcoming Chapel 2.0 release. The standard IO module in particular has undergone some fairly significant improvements and changes, including:

as well as numerous other naming improvements to routines and deprecations of stale functionality.

Better Error Messages via ‘dyno’

Within the compiler’s code base, we have continued making good strides with our ‘dyno’ project, whose goal is to dramatically revamp and modernize Chapel’s compilation architecture to improve compilation times, support separate compilation, etc. Most of these changes are not yet visible to the typical end-user, though one nice one is: A new error-reporting framework has been developed that is now used for all parser errors in the production compiler.

As a result of this change, parse errors now have both a (default) compact message as well as a more detailed one designed to help users cope with the error. To opt in to the latter, compile with the experimental --detailed-errors flag. For example, given the erroneous program:

1
2
3
4

record R {
  var x: int;
}
var x = new R;

the default error message is:

$ chpl newR.chpl
newR.chpl:4: syntax error: 'new' expression is missing its argument list

Meanwhile, recompiling with the new flag gives:

$ chpl newR.chpl --detailed-errors
─── syntax in newR.chpl:4 [NewWithoutArgs] ───
  'new' expression is missing its argument list.
  'new' expression used here:
      |
    4 | var x = new R;
      |
  Perhaps you intended to write 'new R()' instead?

Overload Resolution

Chapel 1.29.0 continues the improvements to overload selection started in Chapel 1.28.0 by preferring procedures that are generic over those that require an implicit conversion to resolve. For example, consider these procedure overloads:

proc f(x)       { writeln("In generic version"); }
proc f(x: real) { writeln("In 'real' version"); }

In Chapel 1.28.0, a call like f(42) would have preferred the real version, but now prefers the generic version, considering it a more precise match. This behavior makes Chapel’s overload selection more similar to C++ as well as more self-consistent with other Chapel cases involving generics.

Notable Bug Fixes

A number of user-identified bugs were fixed in this release, including:

Experimental Features

Turning to experimental features, Chapel 1.29.0 introduces a new ‘weak’ class reference concept for use when working with shared class variables. Though the details of the syntax and interface are still being finalized, this would be a great time to experiment with the feature and offer feedback. See the ‘WeakPointer’ page for details.

The chpl compiler now also supports a prototype capability to capture the generated assembly for a given routine. See ‘Inspecting the Generated Code’ in the LLVM technical note for more information.

In other news, this release includes very experimental support for running Chapel programs on AMD GPUs and for dedicating a core to handling the Chapel runtime’s active messages on Slingshot-11 systems.

For More Information

For a more complete list of changes in Chapel 1.29.0, see its CHANGES.md file. For questions about any of the changes in this release, please reach out to the team on Discourse.

As always, we’re interested in feedback on how we can help make the Chapel language, libraries, implementation, and tools more useful to you in your work.

Thanks to everyone who contributed to Chapel 1.29.0!