v0.19.0

This release now includes a statically linked tarball!
It should work as-is when uncompressed on any amd64 linux system.

Language changes:

• Unused definitions and arguments generate warnings
  • These warnings can be made fatal with --fatal-warnings
• There is now a v0_19_wake stable API package
  • Production wake files should from v0_19_wake import _
  • These files should then continue to work as the API changes

Standard library changes:

• Comparison of NaNs no longer panic, following IEEE rules
• Large Vectors and Integers no longer panic, unlimited size
• Jobs now have an optional Label to control progress display
• New functions: append, prepend, sortPaths

Runtime changes:

• linux systems leverage unshare for isolation
  • This eliminates the '.fuse' directory from build paths
  • Resources fed to the FUSE runner can control unshare
• wake.db is now versioned and detects when incompatible
• When no package is inferred, use an empty package importing wake

Bugfixes:

• Symbol listing ('-g' or '-e') works on large projects
• Unary + and +. are now properly typed
• Missing Tree type was exported
• Work-around for a non-empty .fuse directory added

v0.18.1

Language changes:

• It is illegal to use require where def would suffice.
• It is now legal to use else after a require with a block.

Bugfixes:

• Running wake on a project with excess exports now gives the correct help text.
• Running wake x.y now gives the correct help text.

v0.18.0

The package system:

• Every publish/subscribe topic must now be declared
• Every wake file is now a member of a package
• Packages can export symbols for import into any scope
• Read the docs: https://github.com/sifive/wake/blob/v0.18.0/share/doc/wake/tour/packages.adoc

Language changes:

• Syntax require PATTERN = EXPR [else RETURN-EXPR] to support partial matches
• Anonymous tuples: def (a; b; c) = fnReturnsTriple 42
• Type ascription in expressions is now possible using x : Integer
• Named tuples may no longer be operators

Runtime changes:

• wake now passes its command-line arguments to a function
• executable script support: #! /usr/bin/env wake -:main_function
• auto-initialization of wake.db when a wit-lock.json is found
• Stored expression support was removed
• Support for incremental jobs (eg: invoking make)
• parseJSONBody now fails with Error
• Prevent parsing some .wake files with .wakeignore

Bugfixes:

• Parsing of numbers including _ works again (eg: 0xabcd_1234)
• Changing a Plan's Fn{Inputs,Outputs,Resources} now triggers a rebuild
• Allow up to 128MiB JSON Job descriptions
• Renaming files and directories under FUSE now works reliably
• Fixed ` escapes in regular expressions
• Fixed a few corner cases where wake could segfault
• Prevented unnecessary inlining
• Removed all internal symbol definitions from global/export lists
• Types are now scoped and name conflicts are detected
• Type check failures in pattern matches are reported correctly

v0.17.2

Changes since v0.17.1:

• Backport .wakeignore support from 0.18
• Backport bugfix for when visible <= old inputs
• Forward compatibility with 0.18:
  • Ignore everything on a line starting with the 'topic' keyword
  • Encourage execution of expressions using -x 'EXPR'

v0.17.1

Bugfixes:

• Increase max JSON file size to 128 MiB

v0.15.3

Bugfixes:

• Increase max JSON file size to 128 MiB.

v0.17.0

Language changes:

• none

Runtime changes:

• Now 20x faster than v0.16 on key benchmarks
• Function hashes ignore unused in-scope values
• A spinner now indicates when wake is computing
• New --debug-target diagnoses 'Target subkey mismatch'es
• Developer --no-optimize and --stop-after-ssa options added

Optimizations:

• Uses new SSA form to implement optimization passes
• Interpreter optimized for non-future values
• Heap allocations are reused between GC cycles

Bugfixes:

• JSON parser allocation overrun
• Hash collision on structural sharing
• Commands including newline display properly

v0.15.2

Maintenance release of 0.15 with a few bugfixes:

• FUSE daemon now eventually quits
• FUSE runner now detects rename of files when a directory is renamed
• Don't scan .fuse or .fuse directories for wake files
• Preload runner now detects job output file races
• Wake database now reclaims space incrementally

Wake v0.16.0

Language changes:

• Keyword memoize has been removed
• Plan can specify anticipated memory and CPU Usage
• Multiline strings supported; see the updated quickref
• shellJob runs an inline shell script as a Job
• def Pair x y = ... deconstructor pattern bindings
• Standard library now includes tap* and jfind
• Runner API is now extensible

Runtime changes:

• Garbage collector replaces reference counting (4x speed-up)
• Interactive global progress bar based on prior build stats
• Scheduler understands memory consumption and critical path
• Stack traces include function names and recursion compression
• Database reclaims space incrementally (not rewritten on exit)

Command-line changes:

• -pPERCENT replaces -jJOBS affecting both cores and memory
• --last reports all Jobs run in the last build
• --failed reports all Jobs failed in last build
• --no-tty groups Job command echo + stdout/stderr in logs
• --html creates an interactive reference manual
• --profile-heap reports peak wake memory consumption
• -d --profile=trace.html produces a wake CPU time flame graph

Platform changes:

• New FreeBSD port
• Backported to OSX 10.10
• Backported to Ubuntu 14.04
• RPM packages are now relocatable
• make install/clean now work traditionally

And several dozen bugfixes.

Wake v0.15.1

Language changes:

• None

Runtime changes:

• Hash mismatches are only detected when requested
• Preload runner hides .guard files from jobs
• RegExp escape of ` fixed
• Fixed use of target within a def