codingame_tools.language¶
language
¶
Per-language behavior for CodinGame solutions--file extension, local execution, comment
syntax, and a cg contribution create starter stub--behind one abstract interface,
CgLanguage. This is the only interface outside code should use to access a language; never
import a concrete class like CgPython3Language directly, and never branch on a language ID
string in puzzle_manager/contribution_manager--go through get_language/
get_language_by_extension instead.
Adding a new language is purely additive: drop in a new flat module under
codingame_tools.language.languages (e.g. languages/java.py, exposing a module-level
LANGUAGE: CgLanguage singleton--see codingame_tools.language.registry's module docstring
for the exact discovery contract) and override whichever CgLanguage capabilities it actually
supports. No changes needed anywhere else--codingame_tools.language.registry discovers every
module automatically at load time by walking languages/ (no hardcoded list, no exclusion
list).
codingame_tools.language.default.CgDefaultLanguage is a pure catch-all for a cg_id
CodinGame might add in the future that this client has never seen--every language CodinGame
is confirmed to support today has its own real module under languages/, even one that only
implements extension (see languages/java.py, languages/cpp.py, etc.).
Debugging is per-language, by necessity. A debugger launch is a fundamentally different
mechanism per language, not just a different command: Python runs in-process under debugpy
(codingame_tools.test_runner.debug_stdin), while C++ runs gdb inside a container and is driven
over a pipe by the cpptools adapter (languages/cpp.py). So CgLanguage.start_debug_session
and CgLanguage.vscode_provisioning are per-language rather than shared, and a language gains
debugging by implementing them--see codingame_tools.language.vscode for the generated
launch-configuration contract that keeps them from colliding.
Containerized languages share one image. Anything needing a toolchain cg can't assume is
installed runs in a container built from codingame_tools.language.toolchain's composable
fragments--one image for every language, so a workspace runs one container rather than one per
language. See _docker.py.
DEFAULT_BUILD_TIMEOUT_SECONDS
module-attribute
¶
DEFAULT_BUILD_TIMEOUT_SECONDS = 120.0
Default wall-clock timeout for CgLanguage.build--deliberately far more generous than
DEFAULT_RUN_TIMEOUT_SECONDS, because a cold build can involve pulling/building a container
image and compiling from scratch. Keeping the two separate is the whole reason building is its
own step: a slow first compile must never be reported as a test case timing out.
DEFAULT_RUN_TIMEOUT_SECONDS
module-attribute
¶
DEFAULT_RUN_TIMEOUT_SECONDS = 10.0
Default wall-clock timeout for a single local run--a solution under active development can easily infinite-loop; this keeps a bad run from hanging indefinitely rather than reporting it as a (timed-out) failure.
DEFAULT_TOOLCHAIN_BUILD_TIMEOUT_SECONDS
module-attribute
¶
DEFAULT_TOOLCHAIN_BUILD_TIMEOUT_SECONDS = 3600.0
Default wall-clock timeout for an explicit cg docker toolchain build, as opposed to the
incidental image build DEFAULT_BUILD_TIMEOUT_SECONDS covers.
An order of magnitude more generous because the work is different in kind: composing every supported language downloads a JDK, a .NET SDK and a Node tarball, and pip-builds scientific Python wheels, on a link whose speed cg cannot guess. The run path's 120s is right for "make sure the image is current before this test case"; it would be wrong for "build me the whole thing".
TOOLCHAIN_SUBDIR_NAME
module-attribute
¶
TOOLCHAIN_SUBDIR_NAME = 'docker'
Name of the per-user global toolchain directory under the cg data dir--see
CgLanguageContext.toolchain_dir.
CgBuildProfile
module-attribute
¶
CgBuildProfile = Literal['run', 'debug']
Which flavor of build to produce. "run" is for normal local test execution; "debug" is built
for debuggability instead of speed (no optimization, full symbols) and may compile from a
different source path so a debugger's recorded paths map back to the file the user actually has
open. Interpreted languages ignore this entirely.
CgRunEvent
module-attribute
¶
CgRunEvent = CgRunOutputChunk | CgRunFinished
What CgLanguage.run_streaming() yields: zero or more CgRunOutputChunks as they're produced,
followed by exactly one CgRunFinished.
BASE_IMAGE
module-attribute
¶
BASE_IMAGE = 'debian:bookworm-slim'
The one neutral base every fragment installs onto.
Deliberately not a language image such as gcc:14. A per-language base cannot compose--two of
them cannot both be FROM--and it hides the toolchain version in an image tag, which is exactly
how the C++ build came to be silently two major gcc releases ahead of CodinGame's.
PREAMBLE
module-attribute
¶
PREAMBLE = '# Common to every cg toolchain image, whatever languages it carries.\nENV DEBIAN_FRONTEND=noninteractive\nRUN apt-get update \\\n && apt-get install -y --no-install-recommends ca-certificates coreutils \\\n && rm -rf /var/lib/apt/lists/*\nRUN mkdir -p /opt/cg/env.d /build\nWORKDIR /build\n'
Statements shared by every image, before any fragment.
coreutils supplies the timeout and stdbuf the run and debug paths depend on (already present
on Debian; named so a swapped base still gets them).
CgDockerCleanResult
dataclass
¶
CgDockerCleanResult(containers, images, docker_available)
What clean_managed() tore down.
docker_available
instance-attribute
¶
docker_available
False when Docker isn't installed or the daemon isn't reachable--in which case there was, by definition, nothing to clean, and that isn't an error.
CgDockerError
¶
Bases: Exception
A docker command failed. Carries the command's own output, which is almost always the
actually-useful part.
CgBuildResult
dataclass
¶
CgBuildResult(ok, output, up_to_date)
The outcome of CgLanguage.build.
A result, never an exception, even on failure: a compile error is an expected, routine
outcome that callers need to display (not a crash), and raising would make
cg puzzle play--which does not wrap its loop in a try/except--traceback on a typo.
ok
instance-attribute
¶
ok
Whether a usable artifact now exists. Always True for languages that need no build.
output
instance-attribute
¶
output
Compiler/build diagnostics (warnings even on success, errors on failure). Empty when there was nothing to do. Never contains program output--build is a separate subprocess from run precisely so that build noise can never contaminate a solution's stdout.
up_to_date
instance-attribute
¶
up_to_date
True when nothing had to be rebuilt because the source was unchanged since the last successful build. Lets a caller stay quiet on the common no-op path.
CgDebugSession
dataclass
¶
CgDebugSession(ok, output, details=dict())
A running, ready-to-attach debug target--see CgLanguage.start_debug_session.
ok
instance-attribute
¶
ok
Whether the target actually came up. A result rather than an exception for the same reason
CgBuildResult is: this is driven by an editor's preLaunchTask, where a failed build is a
routine outcome that needs displaying, not a crash.
output
instance-attribute
¶
output
What to show the user--build diagnostics when startup failed, otherwise usually empty.
details
class-attribute
instance-attribute
¶
details = field(default_factory=dict)
Language-specific facts the editor's launch configuration needs, e.g. the container name and
the address gdbserver is listening on. Deliberately untyped: what a debug adapter needs
varies enough per language that a fixed schema would be wrong for the second one.
CgLanguage
¶
CgLanguage(cg_id)
Bases: ABC
A single CodinGame-supported programming language's behavior: how to run a solution locally, its file extension, its single-line-comment syntax, and a starter stub for a freshly-created contribution.
Deliberately has no @abstractmethods: "not supported by this language yet" is the
expected, common state (true for every language but Python3 today, and will stay true
incrementally as languages are added one capability at a time), so every capability below
has a graceful base-class default (raise, for the one genuinely load-bearing operation;
None, for everything else) rather than forcing every new minimal language plugin to
write boilerplate "not implemented" overrides. ABC here is used in the structural/
documentation sense--don't construct this directly; use a language plugin's own singleton
or codingame_tools.language.get_language()/get_language_by_extension().
Source code in codingame_tools/language/base.py
227 228 | |
cg_id
property
¶
cg_id
CodinGame's own canonical identifier for this language, e.g. "Python3", "Java",
"C++"--the exact string used in TestSession/play/TestSession/submit's
programmingLanguageId, and a contribution's solutionLanguage
(createContribution/updateContribution).
extension
property
¶
extension
The file extension (no leading dot, e.g. "py") conventionally used for this
language's solution source, or None if not known. Base implementation: None.
comment_prefix
property
¶
comment_prefix
The single-line-comment prefix for this language's source syntax (e.g. "#" for
Python3), or None if not known. Base implementation: None. See format_comment.
toolchain_fragment
property
¶
toolchain_fragment
This language's contribution to a composed toolchain image, or None if it has no
container support.
Usually a fragment that installs nothing and merely depends on a subsystem plus
supplies its own activation script -- C and C++ both resolve to one gcc and differ only in
whether they export CG_CC or CG_CXX. Installing a toolchain directly here is the
exception, reserved for a language nothing else shares.
Base implementation: None, so a language that is only a name today contributes nothing
to any image rather than silently inflating one.
supports_vscode
property
¶
supports_vscode
Whether build_vscode_provisioning returns anything for this language.
Exists so a caller can tell "already up to date" from "nothing to generate", which are both an empty result. Answering that by calling the builder would need a working directory to build a request from, which a caller reporting an error may not have.
format_comment
¶
format_comment(text)
Format text as a single-line comment in this language's syntax, or None if
comment_prefix isn't known for this language--callers must treat None as "no safe
placeholder text can be generated," not substitute a guessed comment syntax.
Source code in codingame_tools/language/base.py
250 251 252 253 254 255 | |
build_contribution_create_stub_source
async
¶
build_contribution_create_stub_source()
Build a starter data/solution.src for cg contribution create, or None if this
language has no suitable one.
The bar here is "a real, working solution", not "a placeholder", and None is a
correct answer rather than a gap to fill. Contribution/updateContribution validates
server-side: a non-null solutionSource must actually pass every provided test case,
and a contribution must provide at least one local and one validator case. cg
contribution create therefore seeds a trivial pair (input "1" -> output "1"), and
any stub returned here must genuinely satisfy them--Python3's echoes its input for
exactly that reason.
A null solutionSource is explicitly allowed and makes the server skip solution
validation entirely, so returning None keeps push() working: the contribution manager
writes an empty solution.src for it, and sends a blank file as null. Returning a
comment-only placeholder would be strictly worse than None: non-null, failing
validation, and blocking the push. Do not add one here to "fix" a language that returns
None--write a real working solution or leave it.
Contrast format_comment, whose comment-only placeholder is fine for a puzzle: nothing
validates a puzzle's local solution file.
Async so a plugin is free to do real work to produce this (render a template, consult
a language service) rather than only ever returning a fixed string. Base
implementation: None.
Source code in codingame_tools/language/base.py
257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 | |
start_debug_session
async
¶
start_debug_session(ctx, stdin_text, *, timeout=DEFAULT_BUILD_TIMEOUT_SECONDS)
Get the solution ready to be attached to by a debugger, with stdin_text as its stdin,
and return how to reach it.
For a compiled, containerized language this builds the debug profile and starts a stopped
gdbserver; the editor then attaches. Languages whose debugger launches the program
itself (Python3, via debugpy running codingame_tools.puzzle_manager.debug) don't need
this at all and leave it unimplemented.
Redirecting stdin from a file is the whole reason this exists as a separate step: it
lets the redirection happen in a command we control, rather than relying on a debug
adapter's own stdin handling. But the file has to be one the implementation materializes
from stdin_text, not the test case's own file on disk.
That distinction is the entire reason this parameter is text rather than a Path. A
contribution's test-case file carries a final newline this client added (see
common.text_files), so redirecting from it directly would feed the solution one byte
more than cg contribution play does, and one byte more than CodinGame does--verified
2026-08-03 that the server appends nothing. A puzzle's test-case file has no such
addition. Taking text makes the caller resolve that, which it is already positioned to
do, and leaves implementations with one unambiguous job: put exactly these bytes on stdin.
Raises:
-
CgLanguageOperationNotSupportedError–base implementation always raises.
Source code in codingame_tools/language/base.py
284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 | |
stop_debug_session
async
¶
stop_debug_session(ctx)
Tear down whatever start_debug_session started. Idempotent, and safe to call when
nothing is running--it's wired to a postDebugTask, which fires even if the session
never really began. Base implementation: no-op, so a language that needs no teardown
inherits correct behavior.
Source code in codingame_tools/language/base.py
317 318 319 320 321 322 | |
build_vscode_provisioning
async
¶
build_vscode_provisioning(request)
Describe the VS Code run/debug configuration this language wants for the working
directory in request, or None if it has no editor integration yet.
Returns a description; it does not write anything. Where the files go and how they merge
with the user's existing config is codingame_tools.language.vscode's job--a plugin
deliberately has no say in (and no knowledge of) workspace-root resolution.
Base implementation: None.
Source code in codingame_tools/language/base.py
347 348 349 350 351 352 353 354 355 356 | |
build
async
¶
build(ctx, *, profile='run', timeout=DEFAULT_BUILD_TIMEOUT_SECONDS)
Produce whatever artifact run_streaming() needs, if this language needs one at all.
A separate, explicit step rather than something run_streaming() does implicitly, so
that a caller can display build diagnostics separately from program output, report a
compile error once instead of once per test case, and give building its own (much more
generous) timeout. It must be cheap to call repeatedly: a language that compiles is
expected to detect that the source is unchanged since the last successful build and
return up_to_date=True having done nothing.
Base implementation: an immediate no-op success, which is correct for every interpreted language.
Returns:
-
CgBuildResult–A
CgBuildResult. Never raises on a build failure (a compile error is a routine -
CgBuildResult–outcome, not a crash)--check
.ok.
Source code in codingame_tools/language/base.py
358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 | |
run_streaming
¶
run_streaming(ctx, input_text, *, timeout=DEFAULT_RUN_TIMEOUT_SECONDS)
Run the solution described by ctx, feeding input_text to stdin, yielding
CgRunOutputChunks tagged by stream as they're produced and ending with exactly one
CgRunFinished carrying the aggregated CgRunResult. See CgRunOutputChunk for the
stdout/stderr ordering caveat.
Does not build. A language that needs a build artifact expects build() to have been
called first and should fail cleanly if it hasn't.
Raises:
-
CgLanguageOperationNotSupportedError–the base implementation always raises this, immediately (not lazily on iteration); only a language that actually supports local execution overrides it.
Source code in codingame_tools/language/base.py
383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 | |
run
async
¶
run(ctx, input_text, *, timeout=DEFAULT_RUN_TIMEOUT_SECONDS)
Convenience wrapper for a caller that doesn't need progressive output: drains
run_streaming() and returns its final CgRunResult. Not overridden by any language
plugin--every plugin gets this for free once it implements run_streaming().
Raises:
-
CgLanguageOperationNotSupportedError–see
run_streaming().
Source code in codingame_tools/language/base.py
406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 | |
CgLanguageContext
dataclass
¶
CgLanguageContext(root, solution_file, meta_dir, mount_root, toolchain_dir, toolchain_languages=None, toolchain_image=None)
Everything a CgLanguage needs to know about where a solution lives, independent of any
particular run.
Deliberately infallible and identity-free: constructing one must never require a working
directory to have been imported (no reading puzzle.json/contribution.json, no network, no
failure modes). A manager can hand one out for a directory holding nothing but a solution
file. input_text/timeout are deliberately not here--those are per-call, not per-context,
and folding them in would defeat the "build once, run many" split.
root
instance-attribute
¶
root
The puzzle/contribution working directory root (resolved absolute)--the directory holding
data/ and .meta/. Not data/, because .meta/ sits beside it and matters to a build.
solution_file
instance-attribute
¶
solution_file
<root>/data/solution.<ext>--the one real, editable, submittable file, carrying its
language's own extension.
There is exactly one path here, and that is the point. cg used to keep a fixed
data/solution.src with a solution.<ext> symlink beside it, and a debug build had to
choose between them: compiling the link recorded a path the debugger then realpath'd back
to the real file, so the editor navigated away from the file the breakpoints were set in,
and the mapping that fixed navigation broke binding instead. One real file with the right
extension removes the choice.
meta_dir
instance-attribute
¶
meta_dir
The working directory's .meta/ (always <root>/.meta)--gitignored scratch space. Used for
per-root toolchain overrides and generated editor files.
mount_root
instance-attribute
¶
mount_root
The directory a containerized language bind-mounts, at its own path inside the container
(see codingame_tools.language._docker). Normally the VS Code workspace root containing
root, so that in-container paths and host paths are the same string--which is what lets a
generated debug configuration drop sourceFileMap entirely, and lets one container serve
every working directory in the workspace.
Always contains root. Falls back to root itself when there is no enclosing workspace, in
which case a containerized language behaves exactly as it did when it mounted the working
directory.
toolchain_dir
instance-attribute
¶
toolchain_dir
The per-user global toolchain directory (<cg data dir>/docker), holding the shared,
user-tweakable per-language image definitions. Global rather than per-root so that tweaking a
language's toolchain once applies to every puzzle and contribution using that language.
toolchain_languages
class-attribute
instance-attribute
¶
toolchain_languages = None
Languages the container toolchain should carry, or None for every language cg supports.
Defaulted because most callers don't care: the default is the right answer, and a context is constructible without knowing anything about images.
toolchain_image
class-attribute
instance-attribute
¶
toolchain_image = None
A prebuilt toolchain image tag to use instead of composing and building one locally. Skips the Dockerfile entirely -- the point of a published image being a pull rather than a build.
CgLanguageOperationNotSupportedError
¶
CgLanguageOperationNotSupportedError(language, operation)
Bases: Exception
Raised by a CgLanguage method whose base-class default means "not implemented for this
language yet" (currently only run_streaming/run). Callers are expected to catch and
handle this directly--there's no manager-specific translation wrapper.
Source code in codingame_tools/language/base.py
431 432 433 434 | |
CgRunFinished
dataclass
¶
CgRunFinished(result)
The final event yielded by CgLanguage.run_streaming()--every run ends with exactly one
of these, carrying the same aggregated result CgLanguage.run() returns.
CgRunOutputChunk
dataclass
¶
CgRunOutputChunk(stream, text)
A piece of output produced by a running solution, as soon as it's available.
stdout and stderr are two independent, separately-buffered OS pipes--stream says which
one this chunk came from, but the order two chunks from different streams are yielded in
is only the order this reader happened to receive them, not a guarantee about the target
process's true relative write order between the two streams. Treat the two streams as
separate; don't rely on cross-stream ordering.
CgRunResult
dataclass
¶
CgRunResult(output, stderr, returncode, timed_out)
The outcome of running a solution file against one input, once it's finished.
stderr
instance-attribute
¶
stderr
Everything the solution wrote to stderr--not treated as failure by itself (a solution may legitimately write debug output there), but surfaced for inspection when a run does fail.
returncode
instance-attribute
¶
returncode
The subprocess's exit code (0 conventionally means "ran without crashing"). Meaningless
(always -1) when timed_out is True.
timed_out
instance-attribute
¶
timed_out
Whether the run was killed for exceeding its timeout (see DEFAULT_RUN_TIMEOUT_SECONDS).
output/stderr hold whatever was captured before the kill.
CgDefaultLanguage
¶
CgDefaultLanguage(cg_id)
Bases: CgLanguage
No overrides--relies entirely on CgLanguage's base-class defaults (extension unknown,
comment syntax unknown, no stub, no local execution). Used only for a cg_id CodinGame
might add in the future that this client has never seen--every language CodinGame is
confirmed to support today has its own real module under
codingame_tools.language.languages, even one that only implements extension, so it
never falls back to this. Bound to the real cg_id it was looked up with (never a generic
placeholder), so error messages/logging naming .cg_id are always accurate.
Source code in codingame_tools/language/base.py
227 228 | |
CgToolchainError
¶
Bases: Exception
Raised for an unresolvable fragment set--an unknown slug, or a dependency cycle.
CgToolchainFragment
dataclass
¶
CgToolchainFragment(slug, version, depends_on=(), dockerfile='', env_script='')
One composable piece of a toolchain image.
slug
instance-attribute
¶
slug
Stable identifier, used in dependency edges, in the generated header, and as the activation script's filename. Lowercase, no spaces--it appears in shell and Dockerfile contexts.
version
instance-attribute
¶
version
Bumped whenever dockerfile or env_script changes, so an unmodified generated Dockerfile
can be detected as stale and regenerated. Deliberately per fragment rather than one global
template version: changing the Rust fragment shouldn't invalidate a C++-only image.
depends_on
class-attribute
instance-attribute
¶
depends_on = ()
Slugs that must be installed before this one. The mechanism that lets several languages share
a toolchain (C and C++ -> gcc11) and lets conflicting ones coexist (java -> jdk21
while scala -> jvm8).
dockerfile
class-attribute
instance-attribute
¶
dockerfile = ''
Statements inserted verbatim. Legitimately empty: a language whose toolchain is entirely supplied by a subsystem contributes only its dependency edge and its activation script, and emits nothing here. An empty fragment produces no Dockerfile section and so no extra layer.
env_script
class-attribute
instance-attribute
¶
env_script = ''
Body of /opt/cg/env.d/<slug>.sh, if this fragment needs one. The composer prepends the
. <dep>.sh lines itself, so a fragment only writes its own exports.
CgVsCodeMergeError
¶
Bases: Exception
Raised when an existing VS Code config file can't be safely merged into--almost always because it's JSONC (comments/trailing commas) rather than strict JSON. Refusing is deliberate: silently rewriting would drop the user's comments and any content our parser didn't understand.
CgVsCodeProvisioning
dataclass
¶
CgVsCodeProvisioning(configurations=list(), inputs=list(), tasks=list(), files=dict(), retired_names=list(), obsolete_files=list(), recommended_extensions=list())
What a language plugin wants written. Everything is optional--a plugin supplies only the pieces it actually has.
configurations
class-attribute
instance-attribute
¶
configurations = field(default_factory=list)
Entries for launch.json's configurations. Each name must come from entry_name(), so
re-provisioning replaces exactly cg's own entries for this language and leaves everything
else--the user's, and other languages'--alone.
Must not bake in anything specific to request.ctx.root: one entry serves every working
directory in the workspace (see the module docstring).
inputs
class-attribute
instance-attribute
¶
inputs = field(default_factory=list)
Entries for launch.json's inputs. Each id must start with cg_ (see
_OWNED_INPUT_RE), for the same reason.
Nothing populates this today: the pickString test-case pickers it existed for are what the
.meta/-based selection replaced. Kept because it is a real launch.json capability a future
plugin may need, and because write_provisioning must go on pruning 1.0.x leftovers.
tasks
class-attribute
instance-attribute
¶
tasks = field(default_factory=list)
Entries for tasks.json's tasks. Each label must come from entry_name(), and the same
"nothing per-directory" rule applies.
files
class-attribute
instance-attribute
¶
files = field(default_factory=dict)
Extra files to write, keyed by path relative to the working directory root (not the
workspace root)--e.g. .meta/.devcontainer/devcontainer.json.
Belongs under .meta/. These files are generated and not the user's to maintain, and
.meta/ is the only part of a working directory that is gitignored, so anywhere else they'd
be committed into whatever repository tracks the directory. That also rules out data/
specifically, which for a contribution is a git work tree where a stray generated file would
be swept into the server tree by git add -A or deleted by git clean -fd.
retired_names
class-attribute
instance-attribute
¶
retired_names = field(default_factory=list)
Configuration names / task labels earlier versions of this plugin generated and no longer do, so they are removed rather than orphaned.
Mostly a backstop. _is_owned_by_this_run already replaces everything in this language's
namespace, so renaming an action is handled with no declaration at all. This is for the
rarer change that moves an entry out of that namespace--a language's cg_id changing, say--
where nothing else would connect the old name to the new one.
The files equivalent is obsolete_files. Both exist for the same reason: what cg generated
is cg's to clean up, and nothing else's.
obsolete_files
class-attribute
instance-attribute
¶
obsolete_files = field(default_factory=list)
Paths (relative to the working directory root, like files) that earlier versions of this
plugin generated and that should now be deleted.
Generated files are cg's to clean up: a user who upgrades shouldn't be left with a stale
devcontainer.json in a location nothing writes to any more, silently offering VS Code a
"Reopen in Container" that no longer reflects anything. The launch.json equivalent is
_OWNED_NAME_RE.
Only ever names specific files, never directories, and a containing directory is removed only if deleting the file leaves it empty--so a path the user has since put their own work in is left alone.
recommended_extensions
class-attribute
instance-attribute
¶
recommended_extensions = field(default_factory=list)
Extension IDs to merge into extensions.json's recommendations (union, never removing the
user's own--there's no reliable way to tell which ones cg previously added).
CgVsCodeRequest
dataclass
¶
CgVsCodeRequest(ctx, workspace_root, debug_adapter_logging=False)
What a language plugin is being asked to generate configuration for.
Deliberately thin. It used to also carry the working directory's kind and its full list of
test cases, because a generated debug configuration had to name both; now ${file} and
.meta/selected-test.json answer those at launch time, so a plugin needs neither and the
configuration it produces is the same for every working directory in the workspace.
ctx
instance-attribute
¶
ctx
The CgLanguageContext for the working directory (typed loosely to avoid a circular import
with base--see codingame_tools.language.base.CgLanguageContext).
Present for the language's own needs (its toolchain paths, say), not so generated entries can bake in this directory--see the module docstring.
workspace_root
instance-attribute
¶
workspace_root
Where .vscode/ will be written--see find_workspace_root. Often not ctx.root.
Also the directory a containerized language must mount, so that paths inside the container
match the paths VS Code has open--see codingame_tools.language._docker.
debug_adapter_logging
class-attribute
instance-attribute
¶
debug_adapter_logging = False
Generate a configuration that logs the debug adapter's own conversation with the debugger.
Off by default because it is loud and slows a session down. It exists because the debug adapter is the one component of the stack that can't be exercised from a terminal: gdbserver, stdin redirection, stepping and symbol resolution can all be driven by hand and checked, but what VS Code's adapter actually sends and receives can only be observed from inside a real session. When a session misbehaves and everything underneath it demonstrably works, this is the remaining place to look.
A plugin should turn on whatever its adapter offers, and quieten anything so voluminous it would bury the exchange.
build_image_content
async
¶
build_image_content(content, *, tag, platforms=(), push=False, timeout, quiet=False)
Build a Dockerfile given as text, unconditionally--no "is it already built?" check.
Distinct from ensure_image, which is the hot path called before every run and skips the
build whenever the content-addressed tag already exists. This is the explicit
cg docker toolchain build: the user asked for a build, so they get one, and they may ask
for architectures the local daemon can't even load.
Uses the same empty build context as ensure_image (see there): nothing is ever COPY'd
in, because solutions are bind-mounted at run time.
Parameters:
-
platforms(Sequence[str], default:()) –e.g.
["linux/amd64", "linux/arm64"]. Empty means the host's own architecture via plaindocker build. Anything else requires buildx. -
push(bool, default:False) –Push to a registry instead of loading into the local daemon.
Raises:
-
CgDockerError–if buildx is needed but missing, if more than one platform is requested without
push(a hard Docker limitation--see below), or if the build fails.
Source code in codingame_tools/language/_docker.py
439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 | |
clean_managed
async
¶
clean_managed()
Remove every container and image cg created.
Always safe: a container holds nothing but build artifacts, and an image is rebuilt from Dockerfiles that live on disk--so there is no user work here to lose, and everything is recreated on the next build. That's why this neither prompts nor needs a --force.
Containers go first: an image still in use by a container can't be removed.
Source code in codingame_tools/language/_docker.py
612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 | |
compose_dockerfile
¶
compose_dockerfile(directory)
The effective Dockerfile: cg's base with the user's additions appended. Never written to
disk--it's piped straight to docker build.
Source code in codingame_tools/language/_docker.py
326 327 328 329 | |
compose_with_base
¶
compose_with_base(directory, base)
compose_dockerfile, but for a base that has not been written to directory yet.
Exists so a read-only question--"what tag would these languages produce?"--can be answered without the side effect of replacing the user's base.dockerfile. The tag has to cover the composed content, since custom.dockerfile is part of what gets built; computing it from the base alone would report a tag no image ever has.
Source code in codingame_tools/language/_docker.py
332 333 334 335 336 337 338 339 340 341 342 343 344 345 | |
ensure_base_dockerfile
¶
ensure_base_dockerfile(directory, rendered)
Make sure directory holds the current base.dockerfile (and a commented starter
custom.dockerfile), regenerating the base when that's safe.
- Missing -> write it.
- Present, unmodified, but composed from different fragments -> silently regenerate. The normal upgrade path; asks nothing of the user.
- Present and edited -> never overwritten; a stale one produces a warning instead. That
is the escape hatch for swapping
FROMor pinning a toolchain outright.
Parameters:
-
rendered(str) –What the current fragments produce -- see
codingame_tools.language.toolchain.render_dockerfile.
Returns:
-
tuple[Path, list[str]]–The base file's path, and any warnings to surface once.
Source code in codingame_tools/language/_docker.py
271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 | |
ensure_image
async
¶
ensure_image(directory, *, timeout, quiet=False)
Build (or reuse) the toolchain image for the Dockerfiles in directory, returning its tag.
Uses an empty build context (docker build -f - -): nothing is ever COPY'd in, since
the solution is bind-mounted at run time instead. That keeps builds fast and means editing a
solution never invalidates the image.
Docker's own layer cache makes a rebuild of an unchanged Dockerfile nearly free, so this is cheap to call before every build; the image is only genuinely rebuilt when its content- addressed tag changes.
Parameters:
-
quiet(bool, default:False) –Capture
docker buildoutput instead of letting it through to stderr. Off by default because a cold build is slow and its progress is the only feedback.
Source code in codingame_tools/language/_docker.py
357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 | |
image_tag_for
¶
image_tag_for(dockerfile_content)
Content-addressed image tag. Keying on the Dockerfile's content rather than on the working directory means every root sharing the global toolchain shares one image, and any change--a cg template bump or a user tweak--produces a new tag automatically, so nothing ever runs against a stale image.
Source code in codingame_tools/language/_docker.py
220 221 222 223 224 225 | |
latest_alias_for
¶
latest_alias_for()
The stable, non-content-addressed tag kept pointing at the current toolchain image.
Generated devcontainer.json files reference this rather than the real content-addressed
tag: a devcontainer.json is written once and read by VS Code much later, so embedding a hash
that changes on the next toolchain tweak would leave it pointing at an image that no longer
exists. cg's own build/run/debug paths always use the exact tag--this alias exists purely for
tools that need a name stable across rebuilds.
Source code in codingame_tools/language/_docker.py
410 411 412 413 414 415 416 417 418 | |
list_managed_containers
async
¶
list_managed_containers()
Every cg-created container on this machine, as (name, root) pairs.
Found by the cg.root label every cg container carries, so this catches containers for
languages this build doesn't know about and for working directories that no longer exist.
Source code in codingame_tools/language/_docker.py
569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 | |
remove_containers_for_root
async
¶
remove_containers_for_root(root, *, except_name=None)
Remove every cg container bound to root, whatever language--for use when a working
directory is being deleted, and to enforce one-container-per-working-directory.
Matches on the cg.root label rather than recomputing the name, so it catches
containers for languages this build doesn't know about. Silently does nothing when Docker
isn't installed or the daemon is down: failing to tidy up a container must never block
deleting a directory.
Parameters:
-
except_name(str | None, default:None) –Leave this one alone. Used by
ensure_containerto sweep away containers for a working directory's previous language while keeping the current one.
Returns:
-
list[str]–The names removed (empty if none, or if Docker is unavailable).
Source code in codingame_tools/language/_docker.py
639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 | |
tag_image
async
¶
tag_image(source, target)
Point target at the image source already names. Both must be local.
Source code in codingame_tools/language/_docker.py
421 422 423 | |
get_language
¶
get_language(cg_id)
Look up the CgLanguage for a CodinGame protocol language ID.
Always succeeds: returns the real implementation if cg_id matches a discovered language
plugin, or--for a cg_id this client has no record of at all--a CgDefaultLanguage bound
to that ID anyway (memoized, so repeated lookups of the same unrecognized ID return the
same object).
Source code in codingame_tools/language/registry.py
63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 | |
get_language_by_extension
¶
get_language_by_extension(filename_or_extension)
Look up the CgLanguage whose extension matches filename_or_extension (a bare
extension, with or without a leading '.', or a full filename--only the suffix after the
last '.' is considered; case-insensitive).
Returns:
-
CgLanguage | None–The matching
CgLanguage, orNoneif no known language claims this extension--there's -
CgLanguage | None–no reasonable default to fall back to, unlike
get_language.
Source code in codingame_tools/language/registry.py
81 82 83 84 85 86 87 88 89 90 91 92 93 | |
list_language_cg_ids
¶
list_language_cg_ids()
The cg_ids of every discovered language plugin (sorted)--every language CodinGame is
confirmed to support has one, whether or not it implements local execution.
Source code in codingame_tools/language/registry.py
96 97 98 99 | |
all_fragments
¶
all_fragments()
Every fragment cg knows, keyed by slug.
Built fresh rather than cached: the language registry is itself lazily discovered, and a stale copy here would be a second source of truth for something that already has one.
Source code in codingame_tools/language/toolchain/registry.py
27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 | |
default_languages
¶
default_languages()
Every language cg can put in an image -- the default contents of the toolchain.
Derived, never a hardcoded list. A language is in the default set exactly when it declares
a toolchain_fragment, so adding one is a single-module change and the two can never drift
apart.
The default is everything rather than a minimal subset because the whole set costs about
1.9 GB: the languages that dominate (JDK, .NET, Node) share one Debian base instead of each
dragging its own, so trimming saves far less than the confusion of having to choose. Subset
builds remain available for anyone who wants one -- see CgSettingsData.toolchain_languages
-- they are just not something a user should have to think about.
Source code in codingame_tools/language/toolchain/registry.py
51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 | |
fragments_for_languages
¶
fragments_for_languages(languages)
Everything needed to build an image for languages, in install order.
The whole pipeline in one call: names to slugs, slugs to fragments, dependencies pulled in and ordered deterministically.
Source code in codingame_tools/language/toolchain/registry.py
100 101 102 103 104 105 | |
render_dockerfile
¶
render_dockerfile(fragments, *, base_image, preamble='')
The full cg-owned Dockerfile for fragments, already in install order.
Renders cg's own base.dockerfile content. Distinct from
codingame_tools.language._docker.compose_dockerfile, which composes that file on disk with
the user's custom.dockerfile -- generation versus merging.
The header is machine-readable in the same spirit as the single-language one it replaces: it records every fragment and its version, so a generated file can be recognized as cg's, checked for staleness, and told apart from one the user has edited.
Parameters:
-
fragments(list[CgToolchainFragment]) –In install order, as returned by
resolve_fragments. -
base_image(str) –Value for the
CG_BASE_IMAGEbuild arg--one pinned neutral base that every fragment installs onto, rather than a per-language base image. -
preamble(str, default:'') –Statements common to every image, inserted before any fragment.
Source code in codingame_tools/language/toolchain/fragment.py
179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 | |
resolve_language_slugs
¶
resolve_language_slugs(languages)
Fragment slugs for CodinGame language names, e.g. ["C++"] -> ["cpp"].
Raises:
-
CgToolchainError–if a name isn't a known language, or is one with no container support -- distinguished, because "you typed it wrong" and "cg can't containerize that yet" need different fixes.
Source code in codingame_tools/language/toolchain/registry.py
71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 | |
find_workspace_root
¶
find_workspace_root(root)
Best guess at the VS Code workspace root folder that contains root.
Walks up from root looking for a directory that already has .vscode/, then for a VCS
marker (.git/.hg/.svn), stopping at the filesystem root. Falls back to root itself,
which is correct when the user opens the working directory directly as their folder.
This matters because VS Code reads launch.json only from the workspace root--see the
module docstring.
Source code in codingame_tools/language/vscode.py
179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 | |
write_provisioning
¶
write_provisioning(provisioning, *, root, workspace_root, language, force=False, dry_run=False)
Write provisioning into <workspace_root>/.vscode/ (and provisioning.files into
root), merging with whatever is already there.
Only files whose content would actually change are touched, so re-running when everything is already current is a no-op on disk--no diffs, no timestamps, no editor reload prompts.
Parameters:
-
provisioning(CgVsCodeProvisioning) –What the language plugin produced.
-
root(Path) –The working directory root. Only the base for
provisioning.filesnow--entries are owned by declared name, not by a directory. -
workspace_root(Path) –Where
.vscode/lives--seefind_workspace_root. -
language(str) –The
cg_idof the language being provisioned for. Scopes which existing entries this run owns--see_is_owned_by_this_run--so provisioning one language never disturbs another's. -
force(bool, default:False) –Overwrite an existing config file that isn't strict JSON instead of refusing.
-
dry_run(bool, default:False) –Work out what would change without touching anything. See
check_provisioning.
Returns:
-
list[Path]–Every path that changed (or, under
dry_run, would change), in write order. Empty means -
list[Path]–everything was already up to date.
Raises:
-
CgVsCodeMergeError–if an existing file can't be parsed and
forceis False.
Source code in codingame_tools/language/vscode.py
413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 | |