Log every draft command and replay it (t23)

Log every draft command and replay it

Author: peterbjohnson

CHANGELOG.md

@@ -6,4 +6,5 @@
 - `in2lambda FILE FILTER` exits with an error naming the command to run instead, rather than printing its usage and exiting successfully.
 - beartype is now `^0.22`. At 0.20.0 and below its import hook leaves `cli` a plain function rather than a group, so the new command line either fails to import or runs `convert` whatever the arguments; 0.20.1 is the first version that works.
 - `in2lambda source add FILE` freezes a document: it converts .docx and .tex to markdown beside the file, and writes a `draft.json` holding the markdown's hash and every block in it with the lines it spans, so that another tool can quote the source by line range. `in2lambda source show` prints that markdown numbered with the block ids. Freezing a file that has changed since is refused unless `--start-over` says to discard the draft, and so is showing one, since its block ids would name lines they are not the ids of. Both need pandoc and the `convert` extra, as `convert` does.
+- A draft now holds a `log` of every command that changed it and a `fields` map of what those commands wrote, each field recording which layer wrote it (1 a spec, 2 a predicate, 3 a line range, 4 a literal), the source ranges it was copied from, whether it has been edited and by whom. `in2lambda draft mark ignore BLOCK` is the first such command, and `in2lambda draft replay` rebuilds the draft from the frozen markdown and the log, refusing unless what it builds is the `draft.json` that is there, byte for byte. A `draft.json` written before this has no `log` in it and is refused as one nothing here wrote; `in2lambda source add --start-over` freezes the document again.
 - The Python API is unchanged: `in2lambda.main.runner` and everything under `in2lambda.api` take the same arguments and return the same objects.

in2lambda/draft/__init__.py

@@ -0,0 +1,240 @@
+"""Builds up a draft by commands, and rebuilds it from the ones it recorded.
+
+A draft is written by a sequence of commands, some of them chosen by a model. Every
+command that changes one is recorded in the draft's ``log`` as it is applied, and every
+field a command writes carries where it came from, so that :func:`replay` can build the
+same draft again out of the frozen markdown and the log alone, with no model in the loop.
+That is what makes a run reproducible, and a saved run a test.
+
+Commands reach a draft only through :func:`apply`, which is what keeps the log complete:
+a handler registered with :func:`command` is never called by anything else.
+"""
+
+from collections.abc import Callable  # Rather than typing's, which beartype warns on.
+from pathlib import Path
+from typing import Any
+
+from in2lambda.source import (
+    DRAFT,
+    SourceError,
+    _require_conversion_tools,
+    blocks,
+    frozen,
+    save,
+    serialise,
+)
+
+Command = dict[str, Any]
+"""One entry of the log: ``{"command": name, "args": {...}, "by": who}``."""
+
+Handler = Callable[[dict[str, Any], str, dict[str, Any], str], None]
+"""What a command does: `handler(draft, markdown, args, by)`, changing the draft.
+
+The frozen markdown is passed in rather than read, so that a handler quoting the source
+by line range quotes the same text on a replay as it did when it first ran.
+"""
+
+_HANDLERS: dict[str, Handler] = {}
+"""Every command there is, by the name a log entry names it with."""
+
+
+class MalformedCommand(SourceError):
+    """A log holds something that is not a command, so nothing can be made of it."""
+
+
+class UnknownCommand(SourceError):
+    """A log names a command that nothing registered, so the draft cannot be rebuilt."""
+
+
+class NoSuchBlock(SourceError):
+    """A command names a block the frozen source has not got."""
+
+
+class ReplayDiffers(SourceError):
+    """Replaying a draft's log does not reproduce the draft."""
+
+
+def command(name: str) -> Callable[[Handler], Handler]:
+    """Registers a handler as the command of that name.
+
+    Args:
+        name: What a log entry calls it, as it is typed: ``"mark ignore"``.
+
+    Returns:
+        The decorator, which returns the handler unchanged.
+    """
+
+    def register(handler: Handler) -> Handler:
+        _HANDLERS[name] = handler
+        return handler
+
+    return register
+
+
+def record(
+    draft: dict[str, Any],
+    key: str,
+    value: Any,
+    *,
+    layer: int,
+    ranges: list[list[int]],
+    by: str,
+) -> None:
+    """Writes one field of a draft, with where it came from.
+
+    Args:
+        draft: The draft to write into.
+        key: What the field is called, unique within the draft.
+        value: What it is.
+        layer: What wrote it: 1 a spec, 2 a predicate, 3 a range taken from the source,
+            4 a literal someone typed. A reader deciding whether to trust a field wants
+            to know which of those it was.
+        ranges: The line ranges of the frozen source the value was copied from, as
+            ``[[start, end], ...]``, and empty where it was not copied from any.
+        by: Who ran the command, as a name or a model.
+    """
+    draft["fields"][key] = {
+        "value": value,
+        "layer": layer,
+        "ranges": ranges,
+        # A field is edited when something replaces the value a command wrote, which is
+        # not something a command can do to its own field on the way in.
+        "edited": False,
+        "by": by,
+    }
+
+
+def _fault(entry: Any) -> str:
+    """What is wrong with the shape of a log entry, or "" if nothing is."""
+    if not isinstance(entry, dict):
+        return "is not an object"
+    if missing := sorted({"command", "args", "by"} - entry.keys()):
+        return f"has no {' or '.join(missing)}"
+    if not isinstance(entry["command"], str):
+        return f"gives {entry['command']!r} as its command, which is not a name"
+    if not isinstance(entry["args"], dict):
+        return f"gives {entry['args']!r} as its args, which is not an object"
+    return ""
+
+
+def _argument(args: dict[str, Any], name: str, command: str) -> Any:
+    """One argument of a command, given that the log entry gave it.
+
+    Handlers take their arguments through this rather than indexing, so that a log
+    entry missing one says which one rather than raising a KeyError at whoever ran it.
+
+    Raises:
+        MalformedCommand: the entry has no argument of that name.
+    """
+    if name not in args:
+        raise MalformedCommand(
+            f"{args!r} in the log is not a command {command} can run: it has no "
+            f'"{name}" argument.'
+        )
+    return args[name]
+
+
+def apply(draft: dict[str, Any], markdown: str, entry: Any) -> None:
+    """Runs one command against a draft and records it in the draft's log.
+
+    Args:
+        draft: The draft to change, in place.
+        markdown: The frozen markdown the draft was written from.
+        entry: The command, as it is written in the log. Anything at all, rather than a
+            `Command`, because a log is read from a file anyone can edit: what shape it
+            has is something to tell the reader about, not something to assume.
+
+    Raises:
+        MalformedCommand: the entry is not a command.
+        UnknownCommand: nothing is registered under that name.
+    """
+    if fault := _fault(entry):
+        raise MalformedCommand(
+            f"{entry!r} in the log is not a command: it {fault}. A command is an "
+            'object with a "command" naming it, its "args", and who it was run "by".'
+        )
+    if (handler := _HANDLERS.get(entry["command"])) is None:
+        raise UnknownCommand(
+            f"{entry['command']} is not a command this version of in2lambda has, so "
+            "the draft cannot be built from its log. It was written by a newer one."
+        )
+    handler(draft, markdown, entry["args"], entry["by"])
+    # After the handler, so a command that was refused is not recorded as having run.
+    draft["log"].append(entry)
+
+
+def execute(entry: Command, directory: str = ".") -> None:
+    """Runs one command against the draft in a directory and writes it back.
+
+    Args:
+        entry: The command, as it is written in the log.
+        directory: Where the ``draft.json`` to change is.
+
+    Raises:
+        SourceError: the draft is missing, is not one of ours, or was written from
+            markdown that has changed since; or the command is unknown or refused.
+    """
+    draft, markdown = frozen(directory)
+    apply(draft, markdown, entry)
+    save(Path(directory) / DRAFT, draft)
+
+
+def replay(directory: str = ".") -> None:
+    """Rebuilds the draft in a directory from its source and its log, and checks it.
+
+    Nothing is written: the point is to find out whether what is on disk is what its
+    commands say it should be, and a replay that wrote the answer could not tell anyone
+    it was different.
+
+    Args:
+        directory: Where the ``draft.json`` to replay is.
+
+    Raises:
+        DraftExists: the markdown has changed since the draft was written from it, so
+            the commands would be replayed against lines they were not run against.
+        MalformedCommand: the log holds something that is not a command.
+        UnknownCommand: the log names a command nothing here registered.
+        ReplayDiffers: the rebuilt draft is not the one on disk, byte for byte.
+    """
+    _require_conversion_tools()
+    draft, markdown = frozen(directory)
+    # From the markdown rather than from the draft: the blocks are as much a product of
+    # the source as the fields are, and copying them across would not check them.
+    rebuilt: dict[str, Any] = {
+        "source": draft["source"],
+        "hash": draft["hash"],
+        "blocks": [block.to_dict() for block in blocks(markdown)],
+        "log": [],
+        "fields": {},
+    }
+    for entry in draft["log"]:
+        apply(rebuilt, markdown, entry)
+
+    path = Path(directory) / DRAFT
+    if serialise(rebuilt) != path.read_bytes():
+        raise ReplayDiffers(
+            f"Replaying the log in {DRAFT} does not reproduce it, so what is in it did "
+            "not all come from the commands it records - something has changed it since "
+            "they ran. Run in2lambda source add --start-over to begin again."
+        )
+
+
+@command("mark ignore")
+def _mark_ignore(
+    draft: dict[str, Any], markdown: str, args: dict[str, Any], by: str
+) -> None:
+    """Marks one block of the frozen source as nothing to take a question from."""
+    block = _argument(args, "block", "mark ignore")
+    if (found := next((b for b in draft["blocks"] if b["id"] == block), None)) is None:
+        raise NoSuchBlock(
+            f"There is no block {block} in {DRAFT}. Run in2lambda source show to see "
+            "the ids of the blocks there are."
+        )
+    record(
+        draft,
+        f"{block}.ignore",
+        True,
+        layer=3,
+        ranges=[[found["start"], found["end"]]],
+        by=by,
+    )

in2lambda/main.py

@@ -6,18 +6,23 @@
 # import os
 # sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), '..')))
 
+import getpass
 import importlib
 import shlex
+from collections.abc import Iterator  # Rather than typing's, which beartype warns on.
+from contextlib import contextmanager
 from typing import Optional
 
 import rich_click as click
 
+import in2lambda.draft
 import in2lambda.filters
 import in2lambda.source
 from in2lambda.api.set import Set
 
-# Both were defined here before there was an in2lambda.source, and are in other
-# people's scripts as in2lambda.main names.
+# All four are in other people's scripts as in2lambda.main names, whether or not they
+# are used here: `_pandoc` and `file_type` were defined here before there was an
+# in2lambda.source, and `ConversionToolsMissing` is what `runner` documents raising.
 from in2lambda.source import (
     ConversionToolsMissing,
     SourceError,
@@ -27,6 +32,20 @@
 )
 
 
+@contextmanager
+def _message_not_traceback() -> Iterator[None]:
+    """Turns anything raised for a reader into what to do about it and a non-zero exit.
+
+    Every command wraps whatever it calls in this: a missing pandoc, a draft from
+    somewhere else, a source that has moved on are all things the person running it can
+    act on, and none of them are worth a traceback.
+    """
+    try:
+        yield
+    except SourceError as error:
+        raise click.ClickException(str(error)) from None
+
+
 def docx_to_md(docx_file: str) -> str:
     """Converts .docx files to markdown.
 
@@ -188,11 +207,8 @@ def convert(
 ) -> None:
     """Takes in a QUESTION_FILE for a given SUBJECT and produces Lambda Feedback compatible json/zip files."""
     # main() is made separate from click() so that it can be easily imported as part of a library.
-    try:
+    with _message_not_traceback():
         runner(question_file, chosen_filter, output_dir, answer_file)
-    except ConversionToolsMissing as error:
-        # Exit with the install instructions rather than a traceback.
-        raise click.ClickException(str(error)) from None
 
 
 @cli.group("source")
@@ -209,21 +225,49 @@ def source_group() -> None:
 )
 def source_add(file: str, start_over: bool) -> None:
     """Converts FILE to markdown and records its blocks in draft.json beside it."""
-    try:
+    with _message_not_traceback():
         draft = in2lambda.source.add(file, start_over)
-    except SourceError as error:
-        # Exit with what to do about it rather than a traceback.
-        raise click.ClickException(str(error)) from None
     click.echo(f"Wrote {draft}")
 
 
 @source_group.command("show")
 def source_show() -> None:
     """Prints the frozen markdown of the draft in this directory, numbered."""
-    try:
+    with _message_not_traceback():
         click.echo(in2lambda.source.show())
-    except SourceError as error:
-        raise click.ClickException(str(error)) from None
+
+
+@cli.group("draft")
+def draft_group() -> None:
+    """Builds up the draft in this directory, recording every command in it."""
+
+
+@draft_group.group("mark")
+def draft_mark() -> None:
+    """Says what to make of a block of the frozen source."""
+
+
+@draft_mark.command("ignore")
+@click.argument("block")
+@click.option(
+    "--by",
+    default=getpass.getuser,
+    help="Who to record the command as having been run by.  [default: your username]",
+)
+def draft_mark_ignore(block: str, by: str) -> None:
+    """Marks BLOCK as nothing to take a question from."""
+    with _message_not_traceback():
+        in2lambda.draft.execute(
+            {"command": "mark ignore", "args": {"block": block}, "by": by}
+        )
+
+
+@draft_group.command("replay")
+def draft_replay() -> None:
+    """Rebuilds the draft in this directory from its log and checks it is the same."""
+    with _message_not_traceback():
+        in2lambda.draft.replay()
+    click.echo("Replays as it stands.")
 
 
 if __name__ == "__main__":