GitLab

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

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

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

I've rewritten the test suite to be more concise and "correct by
construction".  The primary method that this is achieved by
is introducing the 'TestM' monad, which carries around the
important state for the tests so that (1) we don't have to
pass it as an argument all around, and (2) we can automatically
make the correct decisions about how to do things.  This new
method emphasises "configuration by convention": we assume
that a test-case named "Foo" has its test packages in the
directory "tests/PackageTests/Foo".

A secondary change is that all command functions automatically fail if
they have a non-zero exit code (unless you use the 'shouldFail'
combinator which inverts the sense.)  This saves a lot of typing
on test-cases.  (In fact, I've reorganized all of the commands
related here.)

In the process, I've tightened up the logic for how to find the
LocalBuildInfo of the Cabal we've testing, so we should now
reliably be testing the inplace Cabal library, and not the
system library (as was often the case.)

Because things are a lot shorter, there is no good reason to
make Check modules except for the biggest test cases.  Most
test-cases have been folded into PackageTests.Tests; if you
have a small test-case you should just put it there.
Signed-off-by: Edward Z. Yang <ezyang@cs.stanford.edu>

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

This mode implements #2911, and allows to connect the test runner
directly to stdout/stdin. This is more reliable in the presence of no
threading, i.e. a work-arond for #2398.

I make the test suite use this, so that it passes again, despite
printing lots of stuff. Once #2398 is fixed properly, the test suite
should probably be extended to test all the various --show-details
modes.

The test case is changed to print a rather large amount of text, large
enough to fill a buffer. This way, the buffer draining functionality in
the test runner is tested, and it reveals #2398.

This makes the test suite currently fail, of course, but the bug was
there before.

This gives newcomers like me a better chance to understand how the Cabal
test suite is constructed.

Today in Cabal, when you build and install a package, it is
uniquely identified using an InstalledPackageId which is computed
using the ABI hash of the library that was installed.  There
are few problems with doing it this way:

    - In a Nix-like world, we should instead uniquely identify
      build products by some sort of hash on the inputs to the
      compilation (source files, dependencies, flags).  The ABI
      hash doesn't capture any of this!

    - An InstalledPackageId suggests that we can uniquely identify
      build products by hashing the source and dependencies of
      a package as a whole.  But Cabal packages contain many components:
      a library, test suite, executables, etc.  Currently, when
      we say InstalledPackageId, we are really just talking about
      the dependencies of the library; however, this is unacceptable
      if a Cabal package can install multiple libraries; we need
      different identifiers for each.

    - We've also needed to compute another ID, which we've called
      the "package key", which is to be used for linker symbols
      and type equality GHC-side.  It is confusing what the distinction
      between this ID and InstalledPackageIds are; the main reason
      we needed another ID was because the package key was needed
      prior to compilation, whereas the ABI hash was only available
      afterwards.

This patch replaces InstalledPackageId and PackageKey with a
new identifier called ComponentId, which has the following
properties:

    - It is computed per-component, and consists of a package
      name, package version, hash of the ComponentIds
      of the dependencies it is built against, and the name
      of the component.  For example, "foo-0.1-abcdef" continues
      to identify the library of package foo-0.1, but
      "foo-0.1-123455-foo.exe" would identify the executable,
      and "foo-0.1-abcdef-bar" would identify a private sub-library
      named bar.

    - It is passed to GHC to be used for linker symbols and
      type equality.  So as far as GHC is concerned, this is
      the end-all be-all identifier.

    - Cabal the library has a simple, default routine for computing
      a ComponentId which DOES NOT hash source code;
      in a later patch Duncan is working on, cabal-install can
      specify a more detailed ComponentId for a package
      to be built with.

Here are some knock-on effects:

    - 'id' is a ComponentId

    - 'depends' is now a list of ComponentIds

    - New 'abi' field to record what the ABI of a unit is (as it is no longer
      computed by looking at the output of ghc --abi-hash).

    - The 'HasInstalledPackageId' typeclass is renamed to
      'HasComponentId'.

    - GHC 7.10 has explicit compatibility handling with
      a 'compatPackageKey' (an 'ComponentId') which is
      in a compatible format.  The value of this is read out
      from the 'key' field.
Signed-off-by: Edward Z. Yang <ezyang@cs.stanford.edu>

Prior to this patch, the testsuite used your global package db, and
sometimes, the inplace package db, which led to errors if you didn't
have old-time installed globally. They looked like this:

BuildDeps/SameDepsAllRound:          Cabal result was Result {successful = False, success = Failure, outputText = "\"/home/enolan/cabal/Cabal/tests/Setup configure --user -w  /home/enolan/.nix-profile/bin/ghc\" in PackageTests/BuildDeps/SameDepsAllRound\nConfiguring SameDepsAllRound-0.1...\nSetup: At least the following dependencies are missing:\nold-time -any\n"}

When running the `strip` command (without the -D option) on an installed
.a file, `strip` will write the current time to all file headers in the
.a file. Update the DeterministicAr test to allow this.

The DeterministicAr test is supposed to test for regressions in
relinking avoidance. It is sufficient to do so in the build directory
only.

Fixes #2489. The detailed-0.9 test suite runner would deadlock because
the IO manager was not draining the output pipe of the child
process. Thanks to @r0ml for the patch.

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

In GHC 7.10, Cabal always generate package keys, including in
cases where Backpack was involved (e.g. --instantiated-with).
In fact, in these case, GHC needs to be able to generate the
package key (because it will often make a substitution on the
instantiation, and needs to know if this identity coincides with
anything else we've seen previously).

Thus, we introduce a new notion, the 'LibraryName', which
is JUST the non-Backpack portion of a package key.  For ordinary
packages that are definite, a 'LibraryName' is simply
the 'PackageId' plus 'PackageKey'; for indefinite Backpack packages,
when a package gets instantiatied, it may end up with different
'PackageKey's even though the 'LibraryName' stays the same.
'LibraryName's can be computed purely by Cabal.

This patch:

    - Defines library name, which are the source package ID plus
      a hash of all the source package ID and the library names of external,
      textual dependencies,

    - Redefines the package key to be JUST the hash portion of a
      library name, in the case that Backpack is not used,

    - Records the library name in InstalledPackageInfo.

Note: the source package ID is included both externally (so the library
name is a useful handle to refer to package) and internally (so the
hash can stand alone as the package key.)

A major refactoring which is part of this commit is moving package keys/library
names from LocalBuildInfo to LibComponentBuildInfo.  If you have an LBI, you can
still extract a package key/library name using the new
localPackageKey/localLibraryName function (which looks through the
ComponentBuildInfos of a LocalBuildInfo for the library in question).  This is
conceptually cleaner for two reasons:

    1. Only dependencies of the *library* are counted as part
    of the library name, as opposed to *all* dependencies which
    we previously used.

    2. A library name doesn't really mean much for an executable,
    or a test suite, since no one else will have to link against
    them.  So we can fall back on something simpler.

A more minor refactoring is the 'LibraryName' type, which was
previously defined by LocalBuildInfo and generally looked something
like "HSprocess-0.1-XXXX".  We change the meaning of 'LibraryName'
to be "process-0.1-XXXX" (thus we have to insert some HS additions
in the code) and eliminate componentLibraries, thus assuming that
there is only ONE Haskell library (which was the case.)  So
we remove a little bit of generality and in return get code
that is much easier to read.  (The only downside is GHC's hack
to split DLLs into multiples has to be adjusted slightly, but
this is not a big price to pay.)
Signed-off-by: Edward Z. Yang <ezyang@cs.stanford.edu>

Also fix or silence build warnings

dsf authored Mar 27, 2015 and Duncan Coutts committed Jun 04, 2015

These were dropped during the migration to github.  This patch restores
them, but note that they do not all pass!

Add functionality to allow preprocessors like hsc2hs and C2HS to inform
Cabal of extra C sources that they create that need to be compiled and
linked.  Includes hsc2hs-based test case.

Fixes #2497.

Starting with version 7.10, GHC puts the module interface (.mix) files
for each project in a subdirectory of -hpcdir named for the package
key. We must adjust the search path accordingly when checking for the
.mix file.

- make some paths non-global via / prefix
- remove redundancies
- add /Cabal/Setup and /cabal-install/Setup

Some package tests run multiple tests on the same package, causing the
build directory to be overwritten. For debugging, it is important to
keep the build directory contents, so in this case we run each test
with a different build directory.

It doesn't make any sense to do a profiling build of an executable if
the library doesn't have a profiled build! The option
--enable-executable-profiling is changed to --enable-profiling to
reflect that it is now a global flag. The old option is still recognized
(for now).

Dynamically linking executables will fail without a shared library
unless the executables do not depend on the library or there is no
library. If there is no library, building shared libraries comes at no
cost. If there is a library, the executables usually depend on it, so it
makes sense to --enable-shared.

If the user passes --disable-shared, it will still be honored, but a
warning will be produced.

Also runs the HPC tests regardless of the detected version.

path templates.

CompilerInfo contains more information about the compiler than
CompilerId, which just stores the flavour and version. In particular,
CompilerInfo knows an AbiTag, which can be used to tell binary
incompatible results from the same compiler apart, and detailed
information about supported languages and extensions.

Some fields in CompilerInfo may be left unknown (Nothing). This can
be used in the future to allow partially resolving configurations
based on supported languages or extensions.

getConfigStateFile now throws meaningful exceptions which are caught by
tryGetConfigStateFile and friends, which are allowed to propagate,
rather than just calling 'die'. If the LocalBuildInfo was generated by
an older version of Cabal, an exception is still generated, but the
LocalBuildInfo is included if it is recoverable. This feature is used to
reduce code duplication between the library and the test suite.