GitLab

Trac #10218 reports a subtle bug that turned out to be:

- CSE invalidated the usage information computed
  by earlier demand analysis, by increasing sharing

- that made a single-entry thunk into a multi-entry thunk

- and with -feager-blackholing, that led to <<loop>>

The patch fixes it by making the CSE pass zap usage information for
let-bound identifiers.   It can be restored by -flate-dmd-anal.

(But making -flate-dmd-anal the default needs some careful work;
see Trac #7782.)

The opt_sig production is defined as

  opt_sig :: { ([AddAnn],Maybe (LHsType RdrName)) }
          : {- empty -}                   { ([],Nothing) }
          | '::' sigtype                  { ([mj AnnDcolon $1],Just $2) }

It is used in the alt and decl_no_th productions, but neither of them
add the returned annotations.

This commit captures the annotations in the calling productions.

Reviewed By: austin

Differential Revision: https://phabricator.haskell.org/D822

GHC Trac Issues: #10254

The `guardquals1` production includes

   : guardquals1 ',' qual  {% addAnnotation (gl $ last $ unLoc $1) AnnComma
                                             (gl $2) >>
                               return (sLL $1 $> ($3 : unLoc $1)) }

The AnnComma should be attached to `(gl $ head $ unLoc $1)`, rather than
`last`.

Reviewed By: austin

Differential Revision: https://phabricator.haskell.org/D818

GHC Trac Issues: #10256

The production for opt_kind_sig is

  opt_kind_sig :: { Located (Maybe (LHsKind RdrName)) }
          :                             { noLoc Nothing }
          | '::' kind                   {% ajl (sLL $1 $> (Just $2)) AnnDcolon (gl $1) }

The outer Location is used only to get the full span for the enclosing
declration, and is then stripped. The inner LHsKind then has a SrcSpan that does
not include the '::'

Extend the SrcSpan on $2 to include $1

Reviewed By: austin

Differential Revision: https://phabricator.haskell.org/D813

GHC Trac Issues: #10209

Reviewed By: edsko, austin

Differential Revision: https://phabricator.haskell.org/D824

GHC Trac Issues: #10232

Reviewed By: austin

Differential Revision: https://phabricator.haskell.org/D827

Reviewed By: austin

Differential Revision: https://phabricator.haskell.org/D830

GHC build system have custom autoconf check for `create_timer()`
function from librt. Check description says that it checks for
`CLOCK_REALTIME` timer, but checking code also checks for
`CLOCK_PROCESS_CPUTIME_ID` timer, which is not commonly present (for
example, FreeBSD doesn't have it). This makes whole check fail despite
the fact that FreeBSD have `create_timer()` call and supports
`CLOCK_REALTIME`. As a consequence, GHC RTS falls back to using SIGALRM
for its timing machinery. Not only it's very ancient codepath, it also
break some FFI bindings to C code that isn't prepared for syscall
interruption caused by SIGALRM delivery.

Grepping through ghc source code reveals that `USE_TIMER_CREATE`
defininition in the config.h doesn't imply having
`CLOCK_PROCESS_CPUTIME_ID`. The only place where
`CLOCK_PROCESS_CPUTIME_ID` is used is rts/posix/GetTime.c and this code
handles the absence of `CLOCK_PROCESS_CPUTIME_ID` gracefully.

This patch makes autoconf checking code to check only for
`timer_create(CLOCK_REALTIME, ...)` and fixes check description.

Reviewed By: austin

Differential Revision: https://phabricator.haskell.org/D831

* distrib/configure.ac.in : Detect correct version of LLVM tools as
  was done for the top level configure.ac in 42448e37.
* aclocal.m4 : Add a FIND_LD macro that can be used in both
  configure scripts.
* Use new FIND_LD macro in both configure scripts.
Signed-off-by: Erik de Castro Lopo <erikd@mega-nerd.com>

Test Plan: validate on amd64-linux and armhf-linux.

Reviewers: rwbarton, austin

Subscribers: thomie

Differential Revision: https://phabricator.haskell.org/D828

GHC Trac Issues: #10234

Note [Bind new Givens immediately] in TcRnTypes

We were never using the generality.  Result: less code, more efficient.
Cake for everyone.

Summary: Android has no rand symbol (it's a static inline function there).

Test Plan: ghc-android builds

Reviewers: trofi, austin, hvr

Reviewed By: hvr

Subscribers: thomie

Differential Revision: https://phabricator.haskell.org/D826

GHC Trac Issues: #10274

Test Plan: validate

Reviewers: austin, ezyang, erikd, rwbarton

Reviewed By: ezyang, erikd, rwbarton

Subscribers: erikd, thomie

Differential Revision: https://phabricator.haskell.org/D821

GHC Trac Issues: #10265

Signed-off-by: Edward Z. Yang <ezyang@cs.stanford.edu>

Summary:
We're getting rid of -sig-of foo:A usage, because
it doesn't make sense in any compilation mode besides one-shot,
and we don't expect users to use it anyway.
Signed-off-by: Edward Z. Yang <ezyang@cs.stanford.edu>

Test Plan: validate

Reviewers: simonpj, austin

Subscribers: thomie

Differential Revision: https://phabricator.haskell.org/D789

Signed-off-by: Edward Z. Yang <ezyang@cs.stanford.edu>

Summary:
A common pattern when programming with signatures is to combine multiple
signatures together (signature linking).  We achieve this by making it
not-an-error to have multiple, distinct interface files for the same module
name, as long as they have the same backing implementation.  When a user
imports a module name, they get ALL matching signatures dumped into their
scope.

On the way, I refactored the module finder code, which now distinguishes
between exact finds (when you had a 'Module') and regular finds (when
you had a 'ModuleName').  I also refactored the package finder code to
use a Monoid instance on LookupResult to collect together various results.

ToDo: At the moment, if a signature is declared in the local package,
it completely overrides any remote signatures.  Eventually, we'll want
to also pull in the remote signatures (or even override the local signature,
if the full implementation is available.)  There are bunch of ToDos in the
code for what to do once this is done.

ToDo: At the moment, whenever a module name lookup occurs in GHCi and we
would have seen a signature, we instead continue and return the Module
for the backing implementation.  This is correct for most cases, but there
might be some situations where we want something a little more fine-grained
(e.g. :browse should only list identifiers which are available through
the in-scope signatures, and not ALL of them.)
Signed-off-by: Edward Z. Yang <ezyang@cs.stanford.edu>

Test Plan: validate

Reviewers: simonpj, hvr, austin

Subscribers: carter, thomie

Differential Revision: https://phabricator.haskell.org/D790

GHC Trac Issues: #9252

..due to suppressing base-package module names.

Needs a submodule update on array.

This diff depends on D803.

Reviewed By: austin

Differential Revision: https://phabricator.haskell.org/D806

GHC Trac Issues: #10214

Reviewed By: austin

Differential Revision: https://phabricator.haskell.org/D803

GHC Trac Issues: #10207

Signed-off-by: Edward Z. Yang <ezyang@cs.stanford.edu>

Reviewed By: austin

Differential Revision: https://phabricator.haskell.org/D802

The patch is a last major piece to make unregisterised GHC build
under GCC's link-time optimizer.

Before the patch we imported everything external as functions.
Now we distinct between global variables and functions.

The difference is crucial on ia64 and a complement to fixes:
    > d82f5925
    > CMM: add a mechanism to import C .data labels
    > e18525fa
    > pprC: declare extern cmm primitives as functions, not data
Signed-off-by: Sergei Trofimovich <siarheit@google.com>

Reviewed By: austin

Differential Revision: https://phabricator.haskell.org/D797

In commit 3f30912f an include `ghcconfig.h` was added for
`WORDS_BIGENDIAN`.

Converting the floating point array to a `Word64` array avoids
using the preprocessor altogether and leads to smaller code.

Fixes #10239

Reviewed By: rwbarton, austin

Differential Revision: https://phabricator.haskell.org/D795

GHC Trac Issues: #10239

This fixes Trac #10194

This fixes Trac #10195

For some reason we considered untouchability before, but Trac #10195
shows that this is positively worng.

See Note [Instance and Given overlap] in TcInteract.

Looking at the Git log, it seems that this bug was introduced at the
same time that we introduced the Given/Wanted overlap check in the first
place.

Cosmetic consistency only

We were erroneously quantifying over the function when it
had a dictionary type. A bit pathological, but possible.

This fixes Trac #10251

This fixes Trac #10233

The change is in HscTypes.mkPrintUnqualified, and suppresses the
module qualifier on Names from ghc-prim, base, and template-haskell,
where no ambiguity can aries.  It's somewhat arbitrary, but helps
with things like 'Constraint' which are often not in scope, but
occasionally show up in error messages.

This patch fixes Trac #10148, an outright and egregious
bug in the demand analyser.

It is explained in Note [Demand on case-alternative binders]
in Demand.hs.

I did some other minor refactoring.

To my astonishment I got some big compiler perf changes

* perf/compiler/T5837: bytes allocated -76%
* perf/compiler/T5030: bytes allocated -10%
* perf/compiler/T3294: max bytes used  -25%

Happy days

```
chmod 664 $PATH_TO_GHCI_CONF/.ghci
```

Run ghci. You will now get a warning + a suggestion:

```
  *** WARNING: $PATH_TO_GHCI_CONF/.ghci is writable by someone else, IGNORING!
  Suggested fix: execute 'chmod 644 $PATH_TO_GHCI_CONF/.ghci'
```

Execute fix and the warning should disappear.

Reviewed By: mboes, thomie

Differential Revision: https://phabricator.haskell.org/D805

And mention more prominently that `-DDEBUG` and tests that call
`compiler_stats_num_fields` don't play well together

Reviewed By: austin

Differential Revision: https://phabricator.haskell.org/D799

* don't call `nlHsPar`, as `genOpApp` already does so.
* don't reimplement `isUnboxedTupleTyCon` and `isBoxedTupleTyCon`.

Reviewed By: austin

Differential Revision: https://phabricator.haskell.org/D798