* Hide LaTeX warnings
Currently LaTeX warnings make up about half of our Bazel output which
makes it tricky to find the output that you actually care
about. @bame-da said that he doesn’t care about the warnings so
probably nobody else cares either :)
* Add -halt-on-error
Docker causes a few issues for us here which imho outweigh the
benefits for us atm:
1. It is quite slow since we have to rebuild the image everytime. We
could fix this by caching the image somewhere but even that is
somewhat annoying since Azure does not provide local caches so we
would have to push the image to some external service and fetch it
everytime which is still somewhat slow.
2. It does not work on MacOS (even if you have docker on MacOS it is
not going to work since you will try to install the MacOS release
tarball in a Linux docker container).
On the other hand, DAML assistant allows us to specify the
installation directory and we get a fairly clean environment on CI so
the additional isolation we get from Docker is not that important for us.
* check Foldable.concat doesn't clash with Prelude.concat
* it is not neccessary to use the NoImplicitPrelude ext
* protect against a reasonable (future) optimization from giving a false positive
This was used by the old DAML-LF based record constructor/projection
inliner. We'bve recently replaced by an inlining mechanism which is part
of the conversion from GHC Core to DAML-LF.
* Add mention of cd to setup instructions
* Making instructions unambiguous
* Even clearer instructions
Co-Authored-By: bethaitman-da <42940582+bethaitman-da@users.noreply.github.com>
Without this PR, this variable ends up being set to the string
"variables['System.PullRequest.IsFork']" which meant that we never
uploaded to the Bazel cache.
Rely on zsh features to simplify profile_zsh.sh file.
- Remove `DADE_REPO_ROOT` as it is not needed anywhere, `dade-assist` redefines
it already.
- `${0:A:h}` gives the absolute path of the folder *this* file resides in, no
need for `cd`, `dirname`, and `pwd` combo.
- Use simple process redirection (`<`), as it is more robust than the one in
bash, and works better than `<<<`
We used to have a workaround for the command line length limit on
Windows, which exceeded when building ghc-lib. The ghc-lib package was
recently split into two smaller packages and this workaround is not
needed anymore.
* Update rules_haskell
This updates to the latest rules_haskell, which fixes a few issues on
Windows. Most importantly it flags a few Windows libraries as "system"
libraries, preventing Hazel to fail because they are not provided
through Bazel.
* Fix the streaming-commons build on Windows
This modifies our custom streaming-commons BUILD file to make it work on
Windows. In particular it swaps some system libraries, passes the
`-DWINDOWS` flag and enables the build of an extra module.
* Clean up bazel_tools BUILD file
This wraps a very long line for legibility.
* Fix shared object issues on Windows
* This fixes rules_haskell to use the correct Windows path separator on Windows.
GHC expects the LD_LIBRARY_PATH variable to be a list of semi-colon separated
paths, as opposed to a list of colon separated paths:
51fd357119/compiler/ghci/Linker.hs (L1646-L1650)
* This fixes the name of Haskell shared objects on Windows. By default
Bazel's cc_library generates '.so' files, whereas GHC expects a `.dll`
(or a few other extensions, non of which are `.so`):
51fd357119/rts/linker/PEi386.c (L684)
* Build daml-lf-ast on Windows CI
Allows users to supply the argument -V or --verbosity with a number from 0 to 4 for additional logging.
Also, the first and last log message is logged as a warning, when it really
should just be on INFO level.
* Make daml-asisstant forward --help as intended.
* Refactor command-line parser in daml-assistant.
* Fix issues with daml-assistant CLI.
* Misalign things for Martin.
We add a new test for disjunction choices which uses our custom surface
syntax. It follows the example in the blog post announcing disjunction
choices, which is still under review.
We also remove the old tests for flexible controllers. The test for
DAML-LF 1.1 is not run anymore anyway and the test for DAML-LF 1.2 does not
use our custom surface syntax but the type class implementation underlying
it.