Skip to main content
Version: 2.22

File system

How to safely interact with the file system in your plugin.

It is not safe to use functions like open or the non-pure operations of pathlib.Path like you normally might: this will break caching because they do not hook up to Pants's file watcher.

Instead, Pants has several mechanisms to work with the file system in a safe and concurrent way.

Missing certain file operations?

If it would help you to have a certain file operation, please let us know by either opening a new GitHub issue or by messaging us on Slack in the #plugins room.

Core abstractions: Digest and Snapshot​

The core building block is a Digest, which is a lightweight reference to a set of files known about by the engine.

  • The Digest is only a reference; the files are stored in the engine's persistent content-addressable storage (CAS).
  • The files do not need to actually exist on disk.
  • Every file uses a relative path. This allows the Digest to be passed around in different environments safely, such as running in a temporary directory locally or running through remote execution.
  • The files may be binary files and/or text files.
  • The Digest may refer to 0 - n files. If it's empty, the digest will be equal to pants.engine.fs.EMPTY_DIGEST.
  • You will never create a Digest directly in rules, only in tests. Instead, you get a Digest by using CreateDigest or PathGlobs, or using the output_digest from a Process that you've run.

Most of Pants's operations with the file system either accept a Digest as input or return a Digest. For example, when running a Process, you may provide a Digest as input.

A Snapshot composes a Digest and adds the useful properties files: tuple[str, ...] and dirs: tuple[str, ...], which store the sorted file names and directory names, respectively. For example:

Snapshot(
    digest=Digest(
        fingerprint="21bcd9fcf01cc67e9547b7d931050c1c44d668e7c0eda3b5856aa74ad640098b",
        serialized_bytes_length=162,
    ),
    files=("f.txt", "grandparent/parent/c.txt"),
    dirs=("grandparent", "grandparent/parent"),
)

A Snapshot is useful when you want to know which files a Digest refers to. For example, when running a tool, you might set argv=snapshot.files, and then pass snapshot.digest to the Process so that it has access to those files.

Given a Digest, you may use the engine to enrich it into a Snapshot:

from pants.engine.fs import Digest, Snapshot
from pants.engine.rules import Get, rule

@rule
async def demo(...) -> Foo:
    ...
    snapshot = await Get(Snapshot, Digest, my_digest)

CreateDigest: create new files​

CreateDigest allows you to create a new digest with whichever files you would like, even if they do not exist on disk.

from pants.engine.fs import CreateDigest, Digest, FileContent
from pants.engine.rules import Get, rule

@rule
async def demo(...) -> Foo:
    ...
    digest = await Get(Digest, CreateDigest([FileContent("f1.txt", b"hello world")]))

The CreateDigest constructor expects an iterable including any of these types:

  • FileContent objects, which represent a file to create. It takes a path: str parameter, contents: bytes parameter, and optional is_executable: bool parameter with a default of False.
  • Directory objects, which can be used to create empty directories. It takes a single parameter: path: str. You do not need to use this when creating a file inside a certain directory; this is only to create empty directories.
  • FileEntry objects, which are handles to existing files from DigestEntries. Do not manually create these.

This does not write the Digest to the build root. Use Workspace.write_digest() for that.

PathGlobs: read from filesystem​

PathGlobs allows you to read from the local file system using globbing. That is, sets of filenames with wildcard characters.

from pants.engine.fs import Digest, PathGlobs
from pants.engine.rules import Get, rule

@rule
async def demo(...) -> Foo:
    ...
    digest = await Get(Digest, PathGlobs(["**/*.txt", "!ignore_me.txt"]))
  • All globs must be relative paths, relative to the build root.
  • PathGlobs uses the same syntax as the sources field, which is roughly Git's syntax. Use * for globs over just the current working directory, ** for recursive globs over everything below (at any level the current working directory), and prefix with ! for ignores.
  • PathGlobs will ignore all values from the global option pants_ignore.

By default, the engine will no-op for any globs that are unmatched. If you want to instead warn or error, set glob_match_error_behavior=GlobMatchErrorBehavior.warn or GlobMatchErrorBehavior.error. This will require that you also set description_of_origin, which is a human-friendly description of where the PathGlobs is coming from so that the error message is helpful. For example:

from pants.engine.fs import GlobMatchErrorBehavior, PathGlobs

PathGlobs(
    globs=[shellcheck.options.config],
    glob_match_error_behavior=GlobMatchErrorBehavior.error,
    description_of_origin="the option `--shellcheck-config`",
)

If you set glob_match_error_behavior, you may also want to set conjunction. By default, only one glob must match. If you set conjunction=GlobExpansionConjunction.all_match, then all globs must match or the engine will warn or error. For example, this would fail, even if the config file existed:

from pants.engine.fs import GlobExpansionConjunction, GlobMatchErrorBehavior, PathGlobs

PathGlobs(
    globs=[shellcheck.options.config, "does_not_exist.txt"],
    glob_match_error_behavior=GlobMatchErrorBehavior.error,
    conjunction=GlobExpansionConjunction.all_match,
    description_of_origin="the option `--shellcheck-config`",
)

If you only need to resolve the file names—and don't actually need to use the file content—you can use await Get(Paths, PathGlobs) instead of await Get(Digest, PathGlobs) or await Get(Snapshot, PathGlobs). This will avoid "digesting" the files to the LMDB Store cache as a performance optimization. Paths has two properties: files: tuple[str, ...] and dirs: tuple[str, ...].

from pants.engine.fs import Paths, PathGlobs
from pants.engine.rules import Get, rule

@rule
async def demo(...) -> Foo:
    ...
    paths = await Get(Paths, PathGlobs(["**/*.txt", "!ignore_me.txt"]))
    logger.info(paths.files)

DigestContents: read contents of files​

DigestContents allows you to get the file contents from a Digest.

from pants.engine.fs import Digest, DigestContents
from pants.engine.rules import Get, rule

@rule
async def demo(...) -> Foo:
    ...
    digest_contents = await Get(DigestContents, Digest, my_digest)
    for file_content in digest_contents:
        logger.info(file_content.path)
        logger.info(file_content.content)  # This will be `bytes`.

The result will be a sequence of FileContent objects, which each have a property path: str and a property content: bytes. You may want to call content.decode() to convert to str.

You may not need DigestContents

Only use DigestContents if you need to read and operate on the content of files directly in your rule.

  • If you are running a Process, you only need to pass the Digest as input and that process will be able to read all the files in its environment. If you only need a list of files included in the digest, use Get(Snapshot, Digest).

  • If you just need to manipulate the directory structure of a Digest, such as renaming files, use DigestEntries with CreateDigest or use AddPrefix and RemovePrefix. These avoid reading the file content into memory.

Does not handle empty directories in a Digest

DigestContents does not have a way to represent empty directories in a Digest since it is only a sequence of FileContent objects. That is, passing the FileContent objects to CreateDigest will not result in the original Digest if there were empty directories in that original Digest. Use DigestEntries instead if your rule needs to handle empty directories in a Digest.

DigestEntries: light-weight handles to files​

DigestEntries allows a rule to obtain the filenames (with content digests) and empty directories from a Digest. The value of a DigestEntries is a sequence of FileEntry and Directory objects representing files and empty directories in the Digest, respectively. That sequence can be passed to CreateDigest to recreate the original Digest.

This is useful if you need to manipulate the directory structure of a Digest without actually needing to bring the file contents into memory (which is what occurs if you were to use DigestContents).

from pants.engine.fs import Digest, DigestEntries, Directory, FileEntry
from pants.engine.rules import Get, rule

@rule
async def demo(...) -> Foo:
    ...
    digest_entries = await Get(DigestEntries, Digest, my_digest)
    for entry in digest_entries:
        if isinstance(entry, FileEntry):
            logger.info(entry.path)
            logger.info(entry.file_digest)  # This will be digest of the content.
        elif isinstance(entry, Directory):
            logger.info(f"Empty directory: {entry.path}")

MergeDigests: merge collections of files​

Often, you will need to provide a single Digest somewhere in your plugin—such as the input_digest for a Process—but you may have multiple Digests that you want to use. Use MergeDigests to combine them all into a single Digest.

from pants.engine.fs import Digest, MergeDigests
from pants.engine.rules import Get, rule

@rule
async def demo(...) -> Foo:
    ...
    digest = await Get(
        Digest,
        MergeDigests([downloaded_tool_digest, config_file_digest, source_files_snapshot.digest],
    )
  • It is okay if multiple digests include the same file, so long as they have identical content.
  • If any digests have different content for the same file, the engine will error. Unlike Git, the engine does not attempt to resolve merge conflicts.
  • It is okay if some digests are empty, i.e. EMPTY_DIGEST.

DigestSubset: extract certain files from a Digest​

To get certain files out of a Digest, use DigestSubset.

from pants.engine.fs import Digest, DigestSubset, PathGlobs
from pants.engine.rules import Get, rule

@rule
async def demo(...) -> Foo:
    ...
    new_digest = await Get(
        Digest, DigestSubset(original_digest, PathGlobs(["file1.txt"])
    )

See the section PathGlobs for more details on how the type works.

AddPrefix and RemovePrefix​

Use AddPrefix and RemovePrefix to change the paths of every file in the digest, while keeping the file contents the same.

from pants.engine.fs import AddPrefix, Digest, RemovePrefix
from pants.engine.rules import Get, rule

@rule
async def demo(...) -> Foo:
    ...
    added_prefix = await Get(Digest, AddPrefix(original_digest, "new_prefix/subdir"))
    removed_prefix = await Get(Digest, RemovePrefix(added_prefix, "new_prefix/subdir"))
    assert removed_prefix == original_digest

RemovePrefix will error if it encounters any files that do not have the requested prefix.

Workspace.write_digest(): save to disk​

To write a digest to disk in the build root, request the type Workspace, then use its method .write_digest().

from pants.engine.fs import Workspace
from pants.engine.rules import goal_rule

@goal_rule
async def run_my_goal(..., workspace: Workspace) -> MyGoal:
    ...
    # Note that this is a normal method; we do not use `await Get`.
    workspace.write_digest(digest)
  • The digest will always be written to the build root; you cannot write to arbitrary locations on your machine.
  • You may set the optional parameter path_prefix: str with a relative path.

Workspace is a special type that can only be requested in @goal_rules because it is only safe to write to disk in a @goal_rule. So, a common pattern is for "downstream" rules to return a Digest with the contents they want to write to disk, and then the @goal_rule aggregating all the results and writing them to disk. For example, for the fmt goal, each FmtResult includes a digest field.

For better performance, avoid calling workspace.write_digest multiple times, such as in a for loop. Instead, first, merge all the digests, then write them in a single call.

Bad:

for digest in all_digests:
   workspace.write_digest(digest)

Good:

merged_digest = await Get(Digest, MergeDigests(all_digests))
workspace.write_digest(merged_digest)

DownloadFile​

DownloadFile allows you to download an asset using a GET request.

from pants.engine.fs import DownloadFile, FileDigest
from pants.engine.rules import Get, rule

@rule
async def demo(...) -> Foo:
    ...
    url = "https://github.com/pex-tool/pex/releases/download/v2.1.14/pex"
    file_digest = FileDigest(
        "12937da9ad5ad2c60564aa35cb4b3992ba3cc5ef7efedd44159332873da6fe46",
        2637138
    )
    downloaded = await Get(Digest, DownloadFile(url, file_digest)

DownloadFile expects a url: str parameter pointing to a stable URL for the asset, along with an expected_digest: FileDigest parameter. A FileDigest is like a normal Digest, but represents a single file, rather than a set of files/directories. To determine the expected_digest, manually download the file, then run shasum -a 256 to compute the fingerprint and wc -c to compute the expected length of the downloaded file in bytes.

Often, you will want to download a pre-compiled binary for a tool. When doing this, use ExternalTool instead for help with extracting the binary from the download. See Installing tools.

HTTP requests without digests are unsafe

It is not safe to use DownloadFile for mutable HTTP requests, as it will never ping the server for updates once it is cached. It is also not safe to use the requests library or similar because it will not be cached safely.

You can use a Process with uniquely identifying information in its arguments to run /usr/bin/curl.