The checks might detect some errors in the DSL. The commit also adds non-default
values to some fields to make the checks pass, such as adding a default-language
to each component.

Previously, the solver DSL ignored some types of dependencies when they appeared
in executables or under flags. This commit uses one function, mkBuildInfoTree,
to create all BuildInfos, so that any dependency can be used in any location.

This commit modifies the tests so that they disable executables instead of
libraries.  Disabling executables with 'Buildable: False' is a more realistic
test case.

Allocate different ComponentId for lib and exe.

Do an early check for unbuildable deps, and give good error.

When a library/executable is unbuildable, the solver may still
return a reference to it.  Handle this gracefully in ProjectPlanning
with a nice error message, rather than an assert failure.

Fixes #3978.

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

In the old code, we assumed that only libraries could get
installed, but with executable dependencies this is no
longer true.  Handle it correctly.

This bug caused cycles in the elaborated install plan which
could cause dependency closure to fail, resulting in
weird errors.

Fixes #3996.

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

This reverts commit 992bad44, reversing
changes made to d556ad8a.

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

[ci skip]

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

Add InstallPlan invariant assertion checking, revealing cyclic dep problem

In the old code, we assumed that only libraries could get
installed, but with executable dependencies this is no
longer true.  Handle it correctly.

This bug caused cycles in the elaborated install plan which
could cause dependency closure to fail, resulting in
weird errors.

Fixes #3996.

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

It turns out that the install plan elaboration is constructing cyclic
plans in some cases. The effect is that executing the plan simply misses
out anything that depends on the packages involved in the cycle. This is
probably the cause of #3996

With this patch such cases will fail with an assertion such as:

internal error in InstallPlan.fromSolverInstallPlanWithProgress:
The following packages are involved in a dependency cycle
hspec-discover-2.3.1-da63d0b4e952e7949a113646e4af0aac925a4d864a1db650..

The cause is clearly in the caller of fromSolverInstallPlanWithProgress
and the only caller of that is ProjectPlanning.elaborateInstallPlan.
The problem appears to be to do with the intra-package dependencies.

plus some minor related tidying up, such as fixing a few stray cases
where we still talked about "index" rather than "graph".

For the moment we cannot use ghc-options at the top level in the
cabal.project since it applies to all packages not just local ones.
Until that is fixed, use a workaround of specifying it for the two
local packages of interest directly.

Packages like `alex` include pre-generated lexer/parser source
in their source tarball as e.g.

    dist/build/alex/alex-tmp/Scan.hs
    dist/build/alex/alex-tmp/Parser.hs

to avoid having to have `alex` already installed before building
`alex`...

Unfortunately, 8dc39db7 broke
packages relying on this accidental feature by changing where the
distdir points to. This patch partly reverts that commit and
therefore addresses the regression aspect of #4009.

This fix was suggested by @dcoutts

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

Previously, we unconditionally blasted in all packages, even
if our component didn't actually depend on them.
This fixes #2971.

Also, added a test T2971a which is the opposite problem; previously
we didn't bring in include-dirs from internal libraries.  That
was fixed by the previous commit.

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

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

Add `--dynlibdir`

`--dynlibdir` indicates the directory in which dynamic libraries
are installed. By default this setting is equal to:

`$libdir/$abi`

The static libraries will still end up in:

`$libdir/$libsubdir`

With `$libsubdir/$abi` as the default directory for dynamic
libraries, dynamic libraries will by default end up in a
single shared directory (per package database). This has the
potential to reduce start-up times for dynamically linked
executable as only one RPATH per package database will be
needed.

This commit uses the functionality defined in

https://phabricator.haskell.org/D2611

to tell GHC's > 8.0.1 package database that dynamic libraries
are copied to the directories mentioned in the

`dynamic-library-dirs`

field.

[ci skip]

[ci skip]

[ci skip]

Timer osx pr

Avoid generating redundant constraints in new-freeze

Remove support for --assume-deps-up-to-date

Don't register inplace if we're in per-component mode.

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

We sometimes have to union constraints from different places, but we
should simplify those before rendering so we don't add needless
redundancy like "aeson-pretty ==0.8.1 || ==0.8.1". So we just use
simplifyVersionRange after unioning things.

Fixes #4002

Record the result of a slightly tricky debugging session so the next
person gets a head start.

[skip ci]

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

This reverts commit 351c1306.

This reverts commit 85a76c9d.

This reverts commit 1d27ac71.

This reverts commit 2fac2ec6.

This reverts commit 4ea94c45.

This reverts commit c6c3f2c5.