Module B00.Memo

The type for memoizer feedback. FIXME remove `Miss_tool now that we have notify operations.

The type for memoizers. This ties together an environment, a guard, an operation cache and an executor.

memo is a simpler create

• hash_fun defaults to B00_std.Hash.Xxh_64.
• jobs defaults to B00_std.Os.Cpu.logical_count.
• env defaults to B00_std.Os.Env.current
• cwd defaults to B00_std.Os.Dir:cwd
• cache_dir defaults to Fpath.(cwd / "_b0" / ".cache")
• trash_dir defaults to Fpath.(cwd / "_b0" / ".trash")
• feedback defaults to a nop.

clock m is m's clock.

cpu_clock m is m's cpu clock.

env m is m's environment.

reviver m is m's reviver.

guard m is m's guard.

exec m is m's executors.

trash m is m's trash.

has_failures m is true iff at least one operation (or fiber) has failed.

hash_string m s is B000.Reviver.hash_string (reviver m) s.

hash_file m f is B000.Reviver.hash_file (reviver m) f. Note that these file hashes operations are memoized.

stir ~block m runs the memoizer a bit. If block is true blocks until the memoizer is stuck with no operation and fibers to execute.

status m looks for aggregate errors in m in ops m, see B000.Op.aggregate_error for details.

Usually called after a blocking stir to check everything executed as expected. The function itself has no effect more operations can be on m afterwards. If you are only interested in checking if a failure occured in the memo has_failures is faster.

delete_trash ~block m is B000.Trash.delete ~block (trash m).

ops m is the list of operations that were submitted to the memoizer

mark m is m's mark.

mark m mark is m but operations performed on m are marked by mark.

The type for memoizer fibers returning values of type 'a. A fiber f k represent a thread of execution that eventually kontinues k with its result value when it reaches an end.

Fibers should always be run on a Memo via spawn_fiber or as the continuation of a build operation. A fiber can fail either explictly via fail or because an uncaught exception occurs. In both these cases a `Fail notify operation gets added to the memo to witness the fiber failure. The status of this operation is like any Fail notify: B000.Op.status.Failed.

run m k calls k () asynchronously and handles any fiber failure. This also catches non-asynchronous uncaught exceptions and turns them into `Fail notification operations.

fail m fmt ... fails the fiber via a notify operation.

fail_if_error m r is v if r is Ok v and fail m "%s" e if r is Error _.

Future values.

notify m kind msg is a notification msg of kind kind. Note that a `Fail notification will entail a finish_error, see also fail and fail_if_error.

notify_if_error m kind ~use r is v if r is Ok v. If r is Error e, a notification of kind kind is added to m and use is returned. Note that a `Fail notification will entail a finish_error, see also fail and fail_if_error.

ready m p declares path p to be ready, that is exists and is up-to-date in b. This is typically used with source files and files external to the build (e.g. installed libraries).

read m file k reads the contents of file file as s when it becomes ready and continues with k s.

write m ~reads file w writes file with data w () and mode mode (defaults to 0o644) when reads are ready. w's result must only depend on reads and stamp (defaults to "").

copy m ~mode ?linenum ~src dst copies file src to dst with mode mode (defaults to 0o644) when src is ready. If linenum is specified, the following line number directive is prependend in dst to the contents of src:

#line $(linenum) "$(src)"

mkdir m dir p creates the directory path p with mode mode (defaults to 0o755) and continues with k () whne dir is available. The behaviour with respect to file permission matches Os.Dir.create.

delete m p deletes (trashes in fact) path p and continues with k () when the path p is free to use.

wait_files m files k continues with k () when files become ready. FIXME Unclear whether we really want this.

The type for memoized tools.

The type for memoized tool invocations.

tool m t is tool t memoized. Use the resulting function to spawn the tool with the given arguments.

tool_opt m t is like tool, except None is returned if the tool cannot be found.

spawn m ~reads ~writes ~env ~cwd ~stdin ~stdout ~stderr ~success_exits cmd spawns cmd once reads files are ready and makes files writes ready if the spawn succeeds and the file exists. The rest of the arguments are:

• stdin reads input from the given file. If unspecified reads from the standard input of the program running the build. Warning. The file is not automatically added to reads, this allows for example to use B00_std.Os.File.null.
• stdout and stderr, the redirections for the standard outputs of the command, see stdo. Path to files are created if needed. Warning. File redirections are not automatically added to writes; this allows for example to use B00_std.Os.File.null.
• success_exits the exit codes that determine if the build operation is successful (defaults to 0, use [] to always succeed)
• env, environment variables added to the build environment. This overrides environment variables read by the tool in the build environment except for forced one. It also allows to specify environment that may not be mentioned by the running tool's environment specification.
• cwd the current working directory. Default is cwd. In general it's better to avoid using relative file paths and tweaking the cwd. Construct your paths using the absolute directory functions and make your invocations independent from the cwd.
• post_exec, if specified is called with the build operation after it has been executed or revived. If it was executed this is called before the operation gets recorded. It can be used to define the reads and writes of the operation if they are difficult to find out before hand. Do not access m in that function.
• k, if specified a fiber invoked once the spawn has succesfully executed with the exit code.
• stamp is used for caching if two spawns diff only in their stamp they will cache to different keys. This can be used to memoize tool whose outputs may not entirely depend on the environment, the cli stamp and the the content of read files.

Note. If the tool spawn acts on a sort of "main" file (e.g. a source file) it should be specified as the first element of reads, this is interpreted specially by certain build tracer.

spawn' is like spawn except the actual file paths written by the spawn need not be determined before the spawn. Only the root directory of writes need to be specified via writes_root. After the spawn executes the writes can be determined via the writes function, the returned paths must be absolute and be prefixed by writes_root (defaults to recursively list all the files rootet in writes_root).