* Make stack traces work with cached top level values
For stack traces to work properly in the presence of cached top level values,
we need to cache the stack trace together with the value and restore the
stack trace when we get the value from the cache.
We also need to push location information under the monadic. Otherwise, the
location information will be removed from the continuation stack before the
monadic action is _executed_.
We also change one test case to use `fail` instead of `error` since `fail` is
more tricky than `error` (due to the lazy expression embedding in DAML-LF).
Unfortunately, the test did not work woth `fail` in the past because of the
issues fixed in this PR.
* Explain special cases when pushing location information
* language: append the version to the output dar name by default.
We now by default output foo-1.0.0.dar instead just foo.dar. Also the
maven coordinate default naming got removed.
* fixing integration tests and quickstart.dar occurences
The hie-core tests are flaky in stack. They can fail with the following
error message
```
hie-core: panic! (the 'impossible' happened)\n (GHC version 8.6.5 for x86_64-unknown-linux):
Dynamic linker not initialised
Please report this as a GHC bug: http://www.haskell.org/ghc/reportabug
```
Explicitly initializing the dynamic linker at session startup should
avoid this issue.
This fixes all flakiness in `damlc test` that I was able to
reproduce. Previously, I got it to fail in about 10% of the cases
whereas now I have successfully run tests 200 times under load without
issues.
There were two issues at play here:
1. We run scenarios in separate threads to be able to kill the running
Shake session quickly even if a scenario has an infinite loop or
something like that (there is a timeout but it’s quite long). This
could result in one of those left-over threads trying to issue a
request while we are already trying to shut down.
To fix that, we wait for the concurrency semaphore to be empty before
shutting down.
2. Just waiting for scenario executions is not quite sufficient as
`runAction` does not wait for all rules to finish (we could just use
runActionSync in `damlc test` but I’d rather make this work
properly). While we do wait for all scenario executions to finish
there is one gRPC request in offInterest that we do not wait for:
gcCtxs.
To fix this, I’ve now routed all gRPC requests through the semaphore
which means that we will also wait for these requests to finish (or
prevent them from spawning).
This makes more sense anyway as scenario executions are mostly fairly
cheap requests while things like setting up the context are expensive
so we want to limit their concurrency.
We should make the concurrency limit configurable but I’ll leave that
for a separate PR.
* Extract template instance doc from newtype constructor doc
* Use new ghc-libs
* Remove type application which is not needed anymore
* Update daml-doc output with template instances
* Update hlint version
Apparently, we never had any tests for this, so this PR adds at least
a rudamentary test. The logic for this is rather stupid and easy to
break and might change soon so I’ll hold off on adding more extensive
tests until this works a bit better.
SS.scenarioServiceClient does not just read the actual client from
some IORef, it registers the available gRPC methods. Apparently we
never knew about this or at least I didn’t.
By only doing this once, we should speed things up a bit and this
fixes once of the assertion failures that we have been seing on
shutdown (pthread_mutex_lock(&mu->mutex) == 0 in sync_posix.cc) which
was caused by trying to register a method from another thread after
destroying the channel.
Add a command line option to `damlc inspect` which allows for configuring the
details level of the pretty printed DAML-LF. Right now the only difference is
that levels bigger than 0 print all location information. Level 0, which is
the default, also prints location information top level declarations.
This feature is useful for debugging location information.
* Show function names in stack trace on failing scenario
So far, we've only shown the location of the function but not its name.
Now, we add the name of the function as well.
* Only report progress when client supports it
This fixes an issue that some people encountered when running hie-core
in Emacs with a version of haskell-lsp that does not understand
progress events.
* Fix tests
* More test fixes
* language: fix: compute correct source root when building dar
Previously we just took the base directory of the main file, which is
wrong when the main is deeper down in the module structure.
* hlinting
* reuse moduleImportPaths
I assume the plan was to implement stack traces. I intend to do that as well
but the message type does not fit my approach. Thus, let's remove it first.
* Print stack traces in the scenario on failure
Currently, we only print the last source location, which is not
particularly helpful for debugging. Now, we put all source locations we
encounter during execution on the continuation stack and print them when
a scenario fails. This PR does not print the names of entered functions
or choices. We leave this for a future PR.
* Address Moritz' comments
Otherwise, we can end up retaining references to the old map which
prevent it from being garbage collected.
On a simple testcase that repeatedly opens and closes a module, this
seems to make memory usage constant whereas it was increasing
each time before.
* Add md_instances field to ModuleDoc
* Extract instances
* Add instance fields to class and type docs.
* Implement distributeInstanceDocs
* Use Maybe list instead of list.
* Render instance docs.
* Add instance data in tests and add a golden test.
* Rename to orphan typeclass instances.
* Upgrade ghc-libs supporting generic templates
* Update Proposal and ComposedKey tests to generic template syntax
* Temporarily patch daml-doc test output (will need to fix for template instances)
* Add exports to doc ctx
* Filter out unexported things.
* Added a golden test
* Fix missing constructors.
* Hide unexported class members
* Add rst golden test as well
* Remove unnecessary HIDEs from stdlib.
* lint
* Upgrade haskell-lsp and lsp-test
There have been some fixes upstream that should hopefully mean that we
no longer need to mark the lsp-tests as flaky on Windows. I am having
trouble reproducing the flakiness locally, so let’s see what happens
on CI.
* Also bump stack.yaml
Currently, we pass the flavour of the type constructor around although that
is cheaply computable from the data constructor itself. Thus, don't pass the
flavour around.
For every enum constructor `Foo` we produce a function `$ctor:Foo` whose
value is simply `Foo`. When we convert `Foo` to DAML-LF, we convert to
a call of `$ctor:Foo`. Since enum constructors are always fully applied,
this indirection is useless. Thus, we remove it in this PR.
GHC produces a function `$WFoo` for every record data constructor `Foo`.
The conversion to DAML-LF produces a function `$ctor:Foo` which does exactly
the same. We stop doing this and use the `$WFoo` function instead. Since GHC
does not produce `$WFoo` for newtypes, we need to produce these nevertheless.
I'll leave solving the corresponding issue for variant types to another PR.
Right now, the set of internal functions seems to depend on the DAML-LF
version although it actually doesn't. Since GHC is smart enough this doesn't
bite us but doesn't feel particularly clean either. If the internal functions
ever start to depend on the DAML-LF version again, we should probably put
them into the environment of the conversion to DAML-LF.
The code removed was responsible for injecting DAML-LF primitives for
`create`, `exercise`, etc. for the old template desugaring. These code
paths are not hit anymore with the new template desugaring.
* Only change project root once
As described in #2449, calling withProjectRoot' twice breaks with
relative paths and is also just silly so this PR fixes this by
factoring out the actual logic from init from execInit which does the
project root thingy.
* Add a regression test
* Avoid one indirection for dictionary construction
For every constructor `Foo` we generate a function `$ctor:Foo` which calls
the DAML-LF constructor for `Foo`. This function is useless since GHC already
creates a function `$WFoo` for us which does the same. This PR remove the
`$ctor:...` functions for type class dictionaries.
This makes a bit of a mess in the `Ctors` data type. This will be cleaned up
when all `$ctor:...` functions are finally gone.
* Fix construction of XInstance templates
This is a small step to solve the package name ambiguity problem.
Putting the package version into the package name allows to import a
package two times with different versions.
For example, we print two spaces if there's no annotation between `def` and
the function. This PR fixes this by using the `empty` document instead of the
document containing the empty string. (Please don't ask me why they are
different.)
Comparing FilePaths is really slow so by mapping them to Ints, we can
speed up dependency chasing significantly.
We might want to switch to doing some kind of global hash consing of
file paths at the Shake level but for now, this seems like a nice
improvement while not being too invasive.
This is roughly an ~8s speedup on my testcase.
Until now the test used the manually desugared syntax for the `Iou`
template. Now, we use the proper `template` syntax. The test is also
renamed from `IouDSL` to `ProposalIou` since the DSL would be confusing
after this change. In fact, we're mostly testing that the
instantiation of `Proposal a` with `a = Iou` actually works.
* Remove tests that are obsolete with new template desugaring
The tests were only relevant during the transition phase to the new
template desugaring. Since that desugaring has landed the test don't
provide value anymore and can be deleted.
* Fix super class dictionary detection
* Use tasty-golden for daml-doc's golden tests
With our hand rolled version of golden tests it is pretty painful to update
the golden files when they must change. After this PR it is as simple as
```
> bazel run //compiler/damlc/tests:daml-doc -- --accept
```
* Address indentation issues
* Use diff command line tool for diffing
* Removed new unused dependecy on Diff package
* Ignore carriage returns in diff
This allows us to GC the unsimplified DALF which decreases memory
usage and GC pressure. This does make GenerateRawDalf slightly slower
which could in theory have an effect on IDE performance since we use
the raw DALF for the scenario service. However, I haven’t been able to
measure any regressions (if it does become an issue, we could disable
optimizations completely in the IDE). In fact, things seem to be
slightly faster.
On my testcase max memory usage does seem to go down a bit but not a
lot (3.0GB to 2.7GB but it fluctuates somewhat between runs).
* Rename HaddockParse -> Extract
* Add --qualify-types option.
* More qualified type options
* Update help text
* Update compiler/damlc/lib/DA/Cli/Damlc/Command/Damldoc.hs
Co-Authored-By: Jost Berthold <jost.berthold@digitalasset.com>
* Accidentally left in a testing change.
* Documenting DocCtx fields
* Mark new options internal
* Fix daml build for usernames with spaces
This PR fixes two issues caused by having spaces in your username:
1. On Windows, we need to quote the path to the daml binary in the
batch wrapper (quotes are not valid in usernames, so no need to worry
about escaping them).
2. Invoking ghc-pkg via callCommand broke since shells are
terrible. Luckily, we can easily get away with just using callProcess
here.
* Update compiler/damlc/lib/DA/Cli/Damlc.hs
Co-Authored-By: associahedron <231829+associahedron@users.noreply.github.com>
This PR moves as much work as possible to GetLocatedImports which
contracry to GetDependencyInformation is shared between rules.
It’s still slower than it should be and somewhat messy but at least
it’s slightly faster and imho cleaner than before.
* Make --shake-profiling use the directory based profiling
The single-file based profiling is rather useless in the IDE and I
always found myself having to modify the source to set `profileDir` so
this PR switches the CLI option to control that instead.
* Add --shake-profiling to damlc ide
* Unify rst and md output in damldocs.
* Fix errors
* Define RenderUnwords / RenderIntercalate in terms of RenderConcat
* Delete the old
* Small output format changes and update golden tests
* Fix table rendering and tests.
* Fix golden test.
* Delete more old
* Implement reviewer suggestions
* Optimize mangleIdentifier
I am slightly embarassed by importing Data.Text.Internal and
Data.Text.Array. However, this does make a very measurable difference:
On my testcase it brought down the time encoding modules from 22s to
17s, so it’s a 5s improvement.
I did quickcheck this against the old implementation, so I’m confident
it behaves correctly. I also ran some criterion benchmarks to see the
difference and on its own it’s even more of a difference.
https://gist.github.com/cocreature/822114257227473ecff1638a88f07788
* GHC is stupid :(
* language: put hash in package directories
We put the package id of the main dalf of a package into the directory
names, where we store the files of that package in the package database.
This way we make sure that two equally named packages don't overwrite
their dependencies and files.
This is in preparation for #2326 as well as for splitting hie-core
into a separate repo. Given that, it explicitely avoids using our
dev-env.
We do need to install a few system packages, so for now this uses the
hosted builder so we can do this. Another option would be to just add
those to our builders. I don’t really have a preference either
way. The builds are < 5 minutes so I don’t expect issues from using
the hosted builders.
* Update ghc-libs to use new template desugaring
* Replace old template typeclasses with generic-friendly ones
* New template desugaring doc
* Fix tests
* Fix damldoc tests regarding Archive choice
* Update visualisation code to not traverse master dictionary
* Additional class method stubs to template instance declaration for upgrades
* Increase stack limit for bond trading compilation test
* Update hlint version
* Supporting producing sdist tarballs for the HS ledger bindings
The README.md has an explanation for how you can use this.
This should hopefully allow others to experiment with the bindings.
My benchmark is sadly extremely noisy which makes it hard to figure
out how much of an improvement this is but it does seem to at least be
a minor improvement.
* daml-lf: prepare archive proto for Numeric
* replace DECIMAL by NUMERIC in messages and fields
* add nat kind and nat type
* add builtins CAST_NUMERIC and SHIFT_NUMERIC
* daml-lf: remove new builtins from archive proto
* daml-lf: some more doc about numeric
* daml-lf: fix spec
