**Problem**
Current implementation of testQuick depends on the concept of timestamp,
which probably won't work well with the new consistent analysis store or
the idea of remote caching.
**Solution**
This is a step towards cached testing by making the incrementality hermetic
(no longer depends on the timestamp). Instead this calculates the aggregated
SHA-256 of the class files involved in the test.
The BSP server didn't reset old diagnostic messages sent to BSP clients under
certain circumstances. This commit mitigates this edge case and ensures that
diagnostics for files that previously had compilation problems are properly
reset when fresh diagnostics messages are sent.
The culprit was a mismatch of map keys: Files with problems were sometimes recorded
under an absolute path, but later attempted to be retrieved by virtual path.
`clean` should delete the packed dir. If it does not,
the next `compileIncremental`, which is a cache hit, will see that
the packed dir is already there and will not unpack it.
This follows on from #7470, to include all sources, not just managed and
unmanaged, in the source jar, along with all resources (previously only
unmanaged resources were included).
This means that if, for whatever crazy reason, someone does modify the
`sources` task to include additional sources or filter out sources, rather than
using the managed or unmanaged source mechanisms, their changes will still be
reflected in the source jar.
See also https://github.com/sbt/zinc/pull/1326
This adds a new setting `enableConsistentCompileAnalysis`,
which enables the new "Consistent" Analysis format,
which is faster and more repeatable than the status quo.
This is initialized to `true` by default.
It can be opted out either by the setting or using
`-Dsbt.analysis2024=false`.
When loading a scripted test, sbt creates a jar file and loads it.
The path of the jar file is the same for all the batched tests.
We must prevent the JDK from caching this jar file to force a reload after each test.
Otherwise sbt tries to load the auto-plugins of a previous test and fails.
When a macro was compiled against a new scala-library (say 2.13.15),
the runtime classpath of the Scala compiler should not contain an older
scala-library. Otherwise the macro can cause a NoSuchMethodException
during expansion.
This commit updates scala-library in the scalaInstance to the version
on the projects dependency classpath.
The `csrSameVersions` setting can be used to keep dependencies at the
same version.
By default it's used for Scala artifacts. They need to be kept at the
same version because the compiler / reflect are built with
cross-artifact inlining enabled.
`csrSameVersions := Seq(Set(scala-library, scala-reflect, scala-compiler, scalap))`
Users can make use of the new setting in the following way:
- `csrSameVersions += Set[InclExclRule]("com.corp" % "lib", "com.corp" % "lub")`
- `csrSameVersions += Set[InclExclRule]("com.corp" % "lib-family-*")`
When expanding a macro compiled against a new Scala library, the
runtime classpath of the compiler should not contain an older library.
Otherwise a NoSuchMethodException can occur.
A similar issue is present when running the Scala repl through sbt.
An input line compiled against a new library could fail to run if
the repl's runtime classpath is on an old version.
There are a couple of settings / configs that affect this, summary
below. The change in this PR seems to be the most narrow.
`scalaModuleInfo.value.overrideScalaVersion` in sbt
- affects how sbt to sets coursier's `forceScalaVersion` (see below)
- used by librarymanagement.ivy. If true, add a OverrideScalaMediator
See sbt/sbt#2634. Probably not relevant when using coursier.
`autoScalaLibrary` setting in sbt
- automatically add `scala-library` (or `scala3-library`) as a project
dependency
- also used for `forceScalaVersion` (see below)
`CoursierConfiguration.autoScalaLibrary`
- if `true` then Coursier `ResolutionParams.forceScalaVersion` is set
to to `true`
- initialized by sbt to
`autoScalaLibrary.value &&
!ScalaArtifacts.isScala3(sv) &&
!Classpaths.isScala213(sv) && // added in this commit
scalaModuleInfo.forall(_.overrideScalaVersion)`
coursier `ResolutionParams.forceScalaVersion`
- if true, `scala-library` / `scala-reflect` / `scala-compiler` /
`scalap` are forced to the scala version, not actually resolved
- for Scala 3, the `scala3-library` and `scala3-compiler` versions
are forced
For the details about this PR, please see the blog post https://eed3si9n.com/sbt-remote-cache/.
* Add cache basics
* Refactor Attributed to use StringAttributeMap, which is Map[StringAttributeKey, String]
* Implement disk cache
* Rename Package to Pkg
* Virtualize packageBin
* Use HashedVirtualFileRef for packageBin
* Virtualize compile task
Previously, in sbt 2, the Scripted plugin was not resolving any module
that would provide it with the `ScriptedTests` object, and could
therefore not run scripted tests.
With this patch, `scripted-sbt-redux` will be resolved and added to the
classloader that's used to load `sbt.scriptedtest.ScriptedTests`.
**Problem**
Plugins are topologically sorted, but plugins with equal weigh could
modify the same key via `~=` etc, resulting in different builds
depending on the artifact.
**Solution**
This attempts to fix that by first sorting the selected plugins
by the class name before sorting it topologically.
Problem
-------
Starting Scala 2.13.12, Scala 2 has in-sourced the compiler bridge
implementtion, which hopefully will be kept up to date more than the
ones in Zinc.
Solution
--------
This switches to using the pre-compiled compiler bridge for >=2.13.12.
In addition to ExecuteProgress, this new interface allows builds and plugins to receive events when commands start and finish, including the State before and after each command. It also makes cancellation visible to clients by making the Cancelled type top-level.
Fixes https://github.com/sbt/sbt/issues/7327
**Problem**
In builds with mixed Scala patch versions (like scalameta),
it's possible for a core subproject to be set to the lastest 2.12.x,
but the compiler plugin component is cross published to 2.12.0 etc.
`++ 2.12.0` in this case does not work since sbt 1.7.x onwards requires
the queried Scala version to be listed in `crossScalaVersions`.
**Solution**
This implements sbt 1.6.x-like fallback mechanism,
but instead of using the queried version (e.g. 2.12.0) it will set
the Scala version to one of listed versions that is binary compatible.
- remove unused type params
- use `withFilter` if possible
- use `collectFirst` instead of `collect` and `headOption`
- use `length` instead of `size` if `Array` or `String`
- use `foreach` instead of `map`
Problem
-------
`sbt new` (`sbt init`) hardcodes the templates, which I think is ok,
but without changing much, we can make it extensible.
Solution
--------
This adds two new keys `templateDescriptions` and `templateRunLocal`,
which can customize the behavior for in-house usage etc.
Problem
-------
sbt init menu doesn't pick the right template in the latter half.
Solution
--------
This fixes the mapping between the position and the letter.
If the terminal supports ANSI control sequence,
this displays the template list in an interactive way.
The focused template is rendered reversed,
and arrow key can be used to move the focus up/down.
The sbt-reproducible-build fails if two artifacts point to the same file.
When packaging the artifacts of an sbt plugin,
we copy each files to avoid this issue.
**Problem**
You want to get started with sbt, and you don't know
which template to get started with.
**Solution**
This implements an interactive menu on `sbt new` command
when invoked without an argument to list template candidates.
The first option is `scala/toolkit.local`, which locally creates
an sbt build without calling out to Giter8 (GitHub).
Expose what the incremental compiler is doing behind the scenes. The RunProfiler interface has been part of Zinc for a while, but this allows the build itself, or an Sbt plugin, to hook their own implementation.
We expose a list of such listeners to avoid plugins stepping on each other and replacing an existing listener.
This key has been added in 4061dabf4d but it is only available to Sbt itself. Since ExternalHooks is a Java interface, defined in Zinc for a while and fairly stable, I think this should be safe to do.
My main motivation is to allow installing an InvalidationProfiler from an Sbt plugin, thus gaining access to zinc recompilation decisions. See related PR https://github.com/sbt/zinc/pull/1181
Artifactory rejects the legacy artifacts of sbt plugin.
It is now possible to publish to Artifactory
by turning `sbtPluginPublishLegacyMavenStyle` off.
Problem
-------
In sbt 1, platform cross building is implemented using in the user-land
using `%%%` operator, which clevery handles both Scala cross building
and appending platform suffix like sjs1.
However, in general symbolic `%%%` is confusing, and hard to explain.
Solution
--------
In sbt 2, we should subsume the idea of platform cross building,
so `%%` can act as the current `%%%` operator.
This adds a new setting called `platform`, which defaults to
`Platform.jvm` by default.
When a subprojects sets it to `Platform.sjs1`, `ModuleID`s defined using
`%%` operator will inject the platform suffix `_sjs1` **in addition**
to the Scala binary suffix `_2.13` etc.
Note: Explicit JVM dependencies will now require `.platform(Platform.jvm)`.
See https://eed3si9n.com/simplifying-sbt-with-common-settings/
Problem
-------
The behavior of bare settings is confusing in a multi-project build.
This is partly due to the fact that to use `ThisBuild` scoping
the build user needs to be aware of the task implementation,
and know if the task is already defined at project level.
Solution
--------
This changes the interpretation of the baresettings to be common
settings, which works similar to the way `ThisBuild` behaves in sbt 1.x,
but since this would be a simple append at project-level, it should
work for any tasks or settings.
For an sbt plugin, we publish two POM files, the legacy one, and the
new Maven compatible one. The name of the new POM file contains the sbt
cross-version _2.12_1.0. The format of the new POM file is also slightly
different, because we append the sbt cross-version to all artifactIds of
sbt plugins. Hence Maven can resolve the new sbt plugin POM and its
dependencies.
When resolving an sbt plugin, we first try to resolve the new Maven POM
and if it fails we fallback on the legacy one. When parsing the new POM
format, we remove the sbt cross-version from all artifact IDs so that
there is no mismatch between old and new format of dependencies.
Java diagnostics don't have a pointer but we should report them.
Copied implementation from Bloop to translate the position of an
xsbti.Problem to a BSP range.
Problem
-------
Given val root, currently both `root` and synthetic root gets loaded.
This might be caused by build.sbt being virtualized, and no longer
matching the build root directory.
Solution
--------
For now, comparing the canonical paths seems to fix the issue.
Fixes https://github.com/sbt/sbt/issues/7118
Problem
-------
sbtn 1.8.1 was built using ubuntu-latest, which meant picking up newer
glibc.
Solution
--------
This downgraded the ubuntu machine to build sbtn.
I believe this was just an oversight that it's not marked as true since
sbt can handle `debugSession/start`. This change just ensures that
during the initialization process sbt says it's a `debugProvider` for
the same languages as `runProvider` and `testProvider`. It also
correctly marks the build targets as `canDebug`, unless they are sbt
targets.
Problem
-------
Fixes https://github.com/sbt/sbt/issues/7013
In a shared environment, multiple users will try to create
`/tmp/.sbt/` directory, and fail.
Solution
--------
Append a deterministic hash like `/tmp/.sbt1234ABCD` based
on the user home, so the same directory is used for the given
user, but each user would have a unique runtime directory.
Before
remote cache not found for 0.0.0-7c40144bd1c774e6
After
remote cache artifact not found for org.gontard:sbt-test:0.0.0-7c40144bd1c774e6 Some(cached-compile)
This change is a continuation of the work that was done in
https://github.com/sbt/sbt/pull/6874 to allow the Scala 3 compiler to
forward the `diagnosticCode` of an error as well as the other normal
things. This change in dotty was merged in
https://github.com/lampepfl/dotty/pull/15565 and also backported so it
will be in the 3.2.x series release. This change captures the
`diagnosticCode` and forwards it on via BSP so that tools like Metals
can can use this code.
You can see this in the BSP trace now for a diagnostic:
For example with some code that contains the following:
```scala
val x: Int = "hi"
```
You'll see the code in the BSP diagnostic:
``` "diagnostics": [
{
"range": {
"start": {
"line": 9,
"character": 15
},
"end": {
"line": 9,
"character": 19
}
},
"severity": 1,
"code": "7",
"source": "sbt",
"message": "Found: (\u001b[32m\"hi\"\u001b[0m : String)\nRequired: Int\n\nThe following import might make progress towards fixing the problem:\n\n import sourcecode.Text.generate\n\n"
}
]
```
Co-authored-by: Adrien Piquerez <adrien.piquerez@gmail.com>
Refs: https://github.com/lampepfl/dotty/issues/14904
If we use the ProxyTerminal in the background jobs, the logs
would be spread across different terminals, switching from active
client to active client. We want the logs to stick
to the client that started the job.
A new context is created and closed for each state of the MainLoop.
But the context of the backgroundJob must stay alive.
So we use a context that is owned by the BackgroundJobService.
It creates a new logger for each background job and cleans it when
the job stops.
Problem
-------
Since sbt-doge merger `++ <sv> <command1>` has used binary compatibility
as a test to select subproject, but it causes surprising situations like
sbt/sbt#6915, and it blurs the responsibility of YAML file and build
file as the version specified in the version can override the Scala
version test on local laptop.
Solution
--------
This removes the compatibiliy check (backward-only or otherwise),
and require that `<sv>` match one of `crossScalaVersions` using the new
Semantic Version selector pattern.
Suggested by @SethTisue in https://github.com/sbt/sbt/pull/6894#issuecomment-1168042209
Will show multiple lines when different versions are selected for different subprojects.
When no subprojects have a matching Scala version you will get a
'Switch failed' exception anyway, so in that case there is no
change in behavior.
Problem
-------
There's a bug in ipcsocket cleanup logic that effectively
tries to wipe out /tmp.
Workaround
----------
We should fix the underlying bug, but we can start by
explicitly configuring the temp directories ipcsocket uses.
Ref https://github.com/sbt/sbt/issues/6931
This removes OkHttp dependency.
This also removes an experimental feature to customize HTTP for Ivy called
CustomHttp we added in sbt 1.3.0.
Since LM got switched to Coursier in 1.3.0, I don't think we advertized
CustomHttp.
At some point the watchOnTermination callback stopped working. I'm not
exactly sure how or why that happened but it is fairly straightforward
to restore. The one tricky thing was that the callback has the signature
(Watch.Action, _, _, _) => State, which requires propagating the action
to the failWatch command. The easiest way to do this was to add a
mutable field to the ContinuousState. This is rather ugly and reflects
some poor design choices but a more comprehensive refactor is out of
the scope of this fix.
This commit adds a scripted test that ensures that the callback is
invoked both in the successful and unsuccessful watch cases. In each
case the callback deletes a file and we ensure that the file is indeed
absent after the watch exits.
At some point the watchOnTermination callback stopped working. I'm not
exactly sure how or why that happened but it is fairly straightforward
to restore. The one tricky thing was that the callback has the signature
(Watch.Action, _, _, _) => State, which requires propagating the action
to the failWatch command. The easiest way to do this was to add a
mutable field to the ContinuousState. This is rather ugly and reflects
some poor design choices but a more comprehensive refactor is out of
the scope of this fix.
This commit adds a scripted test that ensures that the callback is
invoked both in the successful and unsuccessful watch cases. In each
case the callback deletes a file and we ensure that the file is indeed
absent after the watch exits.
Before
remote cache not found for 0.0.0-7c40144bd1c774e6
After
remote cache artifact not found for org.gontard:sbt-test:0.0.0-7c40144bd1c774e6 Some(cached-compile)
Fixes#6720
Ref #5994
Problem
-------
Sometimes the compiler returns a fake position such as `<macro>`.
This causes this causes InvalidPathException on Windows if we try
to convert it into NIO path.
Solution
--------
Looks for less-than sign in the VirtualFileRef and skip those.
Since BSP requires the diagnostic info to be associated with
files, we probably can't do much.
Fixes#6592
Problem
-------
On Heroku there's timeout.
Solution
--------
This seems to be coming from supershell closing the executor.
Extend the timeout to 30s.
Migrates TreeView.scala to use Contraband from scala.util.parsing.json,
because this is now deprecated.
The TreeView logic is used in the dependencyBrowseTree task.
`systemOut` notifications are buffered so that they are sent at most
once every 20 millisecond. Other RPC messages are not buffered.
As a consequence, some RPC messages can pass in front of some
systemOut notifications.
That's why `sbt --client run` can exit before it receives all the logs.
In general I think it is safer to maintain the order of all messages.
To do so we can force the flush of systemOut before each RPC message.
Normally scripted tests are forked using the JVM that is running sbt.
If set `scripted / javaHome`, forked using it.
```
scripted / javaHome := Some(file("/path/to/jdk-x.y.z"))
```
Or use `java++` command before scripted.
```
sbt> java++ 11!
sbt> scripted
```
Fixes https://github.com/sbt/sbt/issues/6558
Problem
-------
sbt uses SecurityManager feature of JDK to trap `sys.exit` call during
`run`-like tasks, since we emulate `run` and `console` as function calls.
JDK 17 deprecated SecurityManager and it's printing warnings.
Solution
--------
About 10 years go, `exit` was a convenient way of quitting both Scala
REPL and sbt shell. Scala 2.11 broke this by removing the `Predef.exit`.
We still need to worry about `run` potentially calling `sys.exit`
but that can be handled using fork feature.
In the long-run, it probably is better to be JDK 17 compatible.
Currently crossSbtVersions is incorrectly generating a warning that it
is an unused setting (see https://github.com/sbt/sbt/pull/5153). This
PR fixes this by adding it to the list of excluded lint keys.
Fixes#6571.
Ref #6592
When there's an issue like timeout, currently it throws a
ClassCastException because I made some assumption about the underlying
error. This removes the unnecessary casting.
This fixes the closed channel exception generated when running
sbt -timings help. This bug was introduced in sbt 1.4 where a wrapper
was created (Terminal.scala) around the terminal. This meant that the
shutdown hook which had been used previously was no longer working.
This has been fixed by avoiding the use of the JVM shutdown hook and
instead manually running the thunk at a place in the code where
the programme is still able to respond.
The request of the form buildTarget/* often take a sequence of build
targets as parameter. So far if there is an error on a single build
target, the entire request fails.
This is not the best because the client wants the result of the other
build targets anyway:
For example:
- workspace/buildTargets: if one build target has an invalid Scala
version we still want to import the other ones
- buildTarget/scalacOptions: if a dependency cannot be resolved we still
want to import the build targets that do not depend on it
- buildTarget/scalaMainClasses: if buildTarget does not compile we still
want the main classes of the other targets
...
The change is to respond to BSP requests with the successful
build targets and to ignore the failed ones.
This is implemented the same in Bloop since before BSP in sbt.
In https://github.com/build-server-protocol/build-server-protocol/issues/204,
I made a proposal to also add the failed build targets in the response.
Since build.sbt is compiled/evaluated in `sbt.compiler.Eval`,
this commit introduces a `BuildServerEvalReporter` to redirect
the compiler errors to the BSP clients.
A new `finalReport` method is added in the new `EvalReporter` base class
to reset the old diagnostics.
- Restore old type of `bspWorkspace` key for backwards compat.
Instead, introduce `bspFullWorkspace` that includes the
SBT targets
- Log a warning if the client requests, e.g. `scalaMainClasses`
for a SBT target
- Refactor the code that creates the SBT build targets so it
doesn't depend on `sbtFullWorkspace`.
- Add a setting `bspSbtEnabled` to let the user opt-opt of
SBT target export (e.g. to compensate for a client that does
not yet support them)
What is the problem?
When using remote caching, the resource files are not tracked so if they
have changed, pullRemoteCache will deliver both the old resource file
as well as the changed one.
This is a problem, because it's not the behaviour that our users will
expect and it's not in keeping with the contract of this feature.
Why is this happening?
Zinc, sbt's incremental compiler, keeps track of changes that have
been made. It keeps this in what is called the Analysis file.
However, resource files are not tracked in the Analysis file, so
remote caching is not invalidating the unchanged resource file in
place of the latest version.
What is the solution?
PullRemoteCache deletes all of the resources files. After this,
copyResources is called by PackageBin, which puts the latest
version of the resources back.
If the user runs foo/runMain in a project with multiple main classes,
sbt will still warn the user about their being multiple main classes
even though this is a pointless warning since the user either is running
runMain which requires a main class. The run task is also excluded since
by default it prompts the user with a main class selector. The previous
logic for doing this filtering was bad because it only looked at the
first command in a sequence and couldn't handle the foo/runMain case
since it was looking for an exact match with `run` or `runMain`. This
commit relaxes those restrictions to look at all of the strings in the
command as well as splitting the string to check if the last part of the
key ends in run or runMain. This logic could theoretically be incorrect
if the user wrote an input task that was expecting run or runMain as
user input but even in that case the only consequence would be that they
wouldn't see the multiple main class warning which generally isn't all
the helpful unless you are packaging a jar that expects there to be only
one main class. It seems unlikely that that the user would be running a
custom input task that is both packaging a jar and expecting run or
runMain as input strings.
Fixes#6430.
What is the problem?
As detailed in #6430, the @nowarn annotation was not suppressing
warnings, even after the first attempt to fix this in PR#6431.
This first PR fixed the problem for projects using
enablePlugins(SbtPlugin), but not for those using sbtPlugin := true.
Why is this a valuable problem to solve?
The annotation was not working as users would expect.
What is this solution?
I have moved the scalacOptions change from SbtPlugin.projectSettings
to the scalacOptions in the JvmPlugin settings.
Has this been tested?
Yes, a test has been added. Also, this branch was tested successfully
on the twinagle repo (https://github.com/soundcloud/twinagle/pull/224).
Eugene Yokota
Adrien Piquerez
SlowBrainDude
James Roper
Philippus Baalman
Seth Tisue
Lukas Rytz
Martin Duhem
Tammo Steffens
friendseeker
Aleksandra Zdrojowa
unknown
Martin Duhem
MINKYU LEE
kenji yoshida
Matthias Kurz
Michel Davit
Matthew de Detrich
Iulian Dragos
Julien Richard-Foy
Ondra Pelech
Krzysztof Atłasik
Chris Kipp
Alex Klibisz
Arman Bilge
Jakub Kozłowski
Matthew de Detrich
Nabil Abdel-Hafeez
frosforever
gontard
Devin Fisher
Krzysztof Pado
Jason Pickens
Arnout Engelen
Rui Gonçalves
Kamil Podsiadlo
Ethan Atkins
Kamil Podsiadlo
Rikito Taniguchi
Brice Jaglin
Michał Pałka
naf
Amina Adewusi
David Francoeur
kijuky
kxbmap
Nima Taheri
Igal Tabachnik
Eric Peters
William Narmontas
Jason Zaugg
Samuel CLARENC
Samuel CLARENC
Filip Zybała