Source code for apps.bazel_parser.refinement

from __future__ import annotations

import dataclasses
import enum
import logging

import numpy as np
import pandas

from apps.bazel_parser import repo_graph_data



logger = logging.getLogger(__name__)





class Verbosity(enum.Enum):


    SILENT = "SILENT"



    COUNT = "COUNT"



    LIST = "LIST"




@dataclasses.dataclass


class RefinementConfig:
    """This specifies how to refine a dataframe of nodes to a smaller set.

    We expect node_name to be the index of the dataframe.

    All of these fields are exclusionary.
    """



    name_patterns: list[str]



    class_patterns: list[str]



    class_pattern_to_name_patterns: dict[str, list[str]]




def _show_exclusions(
    pattern: str,
    exclusion: pandas.Series[bool] | np.ndarray[tuple[int]],
    df: pandas.DataFrame,
    verbosity: Verbosity,
) -> None:
    if verbosity == Verbosity.SILENT:
        return
    elif verbosity == Verbosity.COUNT:
        count = len(np.where(exclusion)[0])
        logger.info(f"{pattern} = {count}")
    elif verbosity == Verbosity.LIST:
        logger.info(f"{pattern} ->")
        for node in sorted(df.loc[exclusion].index.tolist()):
            logger.info(f"- {node}")
    else:
        raise ValueError(f"Unhandled verbosity: {verbosity}")




def refine_dataframe(
    df: pandas.DataFrame,
    refinement: RefinementConfig,
    verbosity: Verbosity,
) -> pandas.DataFrame:
    include = np.full(len(df), True, dtype=bool)
    exclude_by_name = []
    for pattern in refinement.name_patterns:
        match = df.index.str.fullmatch(pattern)
        exclude_by_name.append(match)
        include &= ~match
    exclude_by_class = []
    for pattern in refinement.class_patterns:
        cls_match = df["node_class"].str.fullmatch(pattern)
        exclude_by_class.append(cls_match)
        include &= ~cls_match
    exclude_by_class_then_name = {}
    for (
        class_pattern,
        name_patterns,
    ) in refinement.class_pattern_to_name_patterns.items():
        name_exclusions = np.full(len(df), False, dtype=bool)
        for name_pattern in name_patterns:
            name_exclusions |= df.index.str.fullmatch(name_pattern)
        node_cls_match = (
            df["node_class"].str.fullmatch(class_pattern) & name_exclusions
        )
        exclude_by_class_then_name[class_pattern] = node_cls_match
        include &= ~node_cls_match
    for pattern, pat_exclusion in zip(
        refinement.name_patterns, exclude_by_name
    ):
        _show_exclusions(
            pattern=pattern,
            exclusion=pat_exclusion,
            df=df,
            verbosity=verbosity,
        )
    for pattern, pat_cls_exclusion in zip(
        refinement.class_patterns, exclude_by_class
    ):
        _show_exclusions(
            pattern=pattern,
            exclusion=pat_cls_exclusion,
            df=df,
            verbosity=verbosity,
        )
    for pattern, pat_cls_name_exclusion in exclude_by_class_then_name.items():
        _show_exclusions(
            pattern=pattern,
            exclusion=pat_cls_name_exclusion,
            df=df,
            verbosity=verbosity,
        )
    return df.loc[include]





def remove_node_from_repo(
    node: str, repo: repo_graph_data.RepoGraphData
) -> None:
    """Modify graph by removing node, but preserving edges.

    Removed nodes are expected to be noise/artifact targets (probability=1.0,
    duration=0), so dropping their attributes is correct. Callers should call
    repo.refresh() afterward to recompute all metrics on the reduced graph.
    """
    for parent in repo.graph.predecessors(node):
        for child in repo.graph.successors(node):
            repo.graph.add_edge(parent, child)
    repo.graph.remove_node(node)





def full_refinement(
    repo: repo_graph_data.RepoGraphData,
    refinement: RefinementConfig,
    verbosity: Verbosity,
) -> pandas.DataFrame:
    refined_df = refine_dataframe(
        df=repo.df,
        refinement=refinement,
        verbosity=verbosity,
    )
    removed_nodes = set(repo.df.index.tolist()) - set(
        refined_df.index.tolist()
    )
    repo.df = refined_df
    for node in removed_nodes:
        remove_node_from_repo(node, repo)
    return repo.df