On my laptop these changes speed up the full build of the RGL and example
grammars with 'cabal build' from ~95s to ~43s and the zero build from ~18s
to ~5s.
The main change is the introduction of the module GF.CompileInParallel that
replaces GF.Compile and the function GF.Compile.ReadFiles.getAllFiles. At
present, it is activated with the new -j flag, and it is only used when
combined with --make or --batch. In addition, to get parallel computations,
you need to add GHC run-time flags, e.g., +RTS -N -A20M -RTS, to the command
line.
The Setup.hs script has been modified to pass the appropriate flags to GF
for parallel compilation when compiling the RGL and example grammars, but you
need a recent version of Cabal for this to work (probably >=1.20).
Some additonal refactoring were made during this work. A new monad is used to
avoid warnings/error messages from different modules to be intertwined when
compiling in parallel, so some functios that were hardiwred to the IO or IOE
monads have been lifted to work in arbitrary monads that are instances in
the appropriate classes.
In particular, the function compileOne has been moved to the new module
GF.CompileOne and its type has been changed from
compileOne :: ... -> CompileEnv -> FilePath -> IOE CompileEnv
to
compileOne :: ... -> SourceGrammar -> FilePath -> IOE OneCompiledModule
making it more suitable for use in a parallel compiler.
The def rules are now compiled to byte code by the compiler and then to
native code by the JIT compiler in the runtime. Not all constructions
are implemented yet. The partial implementation is now in the repository
but it is not activated by default since this requires changes in the
PGF format. I will enable it only after it is complete.
GF.Text.Pretty provides the class Pretty and overloaded versions of the pretty
printing combinators in Text.PrettyPrint, allowing pretty printable values to
be used directly instead of first having to convert them to Doc with functions
like text, int, char and ppIdent. Some modules have been converted to use
GF.Text.Pretty, but not all. Precedences could be added to simplify the pretty
printers for terms and patterns.
GF.Infra.Location contains the types Location and L, factored out from
GF.Grammar.Grammar, and the class HasSourcePath. This allowed the import
of GF.Grammar.Grammar to be removed from GF.Infra.CheckM, making it more
like a pure library module.
PGF exports the public, stable API.
PGF.Internal exports additional things needed in the GF compiler & shell,
including the nonstardard version of Data.Binary.
(table { p_i => t_i } ! x).l ==> table { p_i => t_i.l } ! x
This was used in the old partial evaluator and can significantly reduce term
sizes in some cases.
Eta expansion is applied between partial evaluation and PMCFG generation.
The buggy version generated type incorrect terms, but PMCFG generation
apparently worked anyway.
+ The current type checker for concrete syntax is in
GF.Compile.TypeCheck.RConcrete, but GF.Compile.TypeCheck.Concrete was
still imported in GFI.
+ Fixed a bug that allowed Ints n as a subtype of Ints m, regardless of
m and n. It now requires n<=m. Note: the type checker still allows Int
as a subtype of Ints m, regardless of m.
+ Fixed a potential efficiency problem with large record types, by reducing
the number of recursive calls from |R|*|S| to |R| when checking if R<=S.
+ Fixed a misleading comment: "alpha g t u" checks that u is a subtype of t,
the other way around. Similarly, "checkIfEqLType gr g t u trm" checks that
u is a subtype of t, not the other way around, and not that t is equal to u.
This bug was introduced sometime between 2013-08-21 and 2013-11-01 and caused
the function convertTerm in GF.Compile.GeneratePMCFG to encounter a EPatt where
it expected Strs. I fixed it by applying the function getPatts (from the old
partial evaluator) to the pattern.
This means that the -old-comp and -new-comp flags are not recognized anymore.
The only functional difference is that printnames were still normalized with
the old partial evaluator. Now that is done with the new partial evaluator.
This was a fairly simple change thanks to previous work on making the Ident
type abstract and the fact that PGF.CId already uses UTF-8-encoded
ByteStrings.
One potential pitfall is that Data.ByteString.UTF8 uses the same type for
ByteStrings as Data.ByteString. I renamed ident2bs to ident2utf8 and
bsCId to utf8CId, to make it clearer that they work with UTF-8-encoded
ByteStrings.
Since both the compiler input and identifiers are now UTF-8-encoded
ByteStrings, the lexer now creates identifiers without copying any characters.
**END OF DESCRIPTION***
Place the long patch description above the ***END OF DESCRIPTION*** marker.
The first line of this file will be the patch name.
This patch contains the following changes:
M ./src/compiler/GF/Compile/CheckGrammar.hs -3 +3
M ./src/compiler/GF/Compile/GrammarToPGF.hs -2 +2
M ./src/compiler/GF/Grammar/Binary.hs -5 +1
M ./src/compiler/GF/Grammar/Lexer.x -11 +13
M ./src/compiler/GF/Infra/Ident.hs -19 +36
M ./src/runtime/haskell/PGF.hs -1 +1
M ./src/runtime/haskell/PGF/CId.hs -2 +3
1. The default encoding is changed from Latin-1 to UTF-8.
2. Alternate encodings should be specified as "--# -coding=enc", the old
"flags coding=enc" declarations have no effect but are still checked for
consistency.
3. A transitional warning is generated for files that contain non-ASCII
characters without specifying a character encoding:
"Warning: default encoding has changed from Latin-1 to UTF-8"
4. Conversion to Unicode is now done *before* lexing. This makes it possible
to allow arbitrary Unicode characters in identifiers. But identifiers are
still stored as ByteStrings, so they are limited to Latin-1 characters
for now.
5. Lexer.hs is no longer part of the repository. We now generate the lexer
from Lexer.x with alex>=3. Some workarounds for bugs in alex-3.0 were
needed. These bugs might already be fixed in newer versions of alex, but
we should be compatible with what is shipped in the Haskell Platform.
+ Eliminated vairous ad-hoc coersion functions between specific monads
(IO, Err, IOE, Check) in favor of more general lifting functions
(liftIO, liftErr).
+ Generalized many basic monadic operations from specific monads to
arbitrary monads in the appropriate class (MonadIO and/or ErrorMonad),
thereby completely eliminating the need for lifting functions in lots
of places.
This can be considered a small step forward towards a cleaner
compiler API and more malleable compiler code in general.
Add proper type checking of course-of-values tables:
+ Make sure that all subterms have the same type.
+ Resolve overloaded operators.
Note though that the GF book states in C.4.12 that the "course-of-values
table [...] format is not recommended for GF source code, since the
ordering of parameter values is not specified and therefore a
compiler-internal decision."
This module should not be part of the public PGF library API, and it was only
used in GF.CompileToAPI, so the code was moved there. The module defined
constFuncs and syntaxFuncs, but only syntaxFuncs was used.
+ References to modules under src/compiler have been eliminated from the PGF
library (under src/runtime/haskell). Only two functions had to be moved (from
GF.Data.Utilities to PGF.Utilities) to make this possible, other apparent
dependencies turned out to be vacuous.
+ In gf.cabal, the GF executable no longer directly depends on the PGF library
source directory, but only on the exposed library modules. This means that
there is less duplication in gf.cabal and that the 30 modules in the
PGF library will no longer be compiled twice while building GF.
To make this possible, additional PGF library modules have been exposed, even
though they should probably be considered for internal use only. They could
be collected in a PGF.Internal module, or marked as "unstable", to make
this explicit.
+ Also, by using the -fwarn-unused-imports flag, ~220 redundant imports were
found and removed, reducing the total number of imports by ~15%.
