[llvm] [Dexter] Remove clang-opt-bisect analyzer tool (PR #128165)
Jeremy Morse via llvm-commits
llvm-commits at lists.llvm.org
Fri Feb 21 03:26:25 PST 2025
https://github.com/jmorse created https://github.com/llvm/llvm-project/pull/128165
The idea behind this tool was that you'd instrument a source file to measure its debug-info quality, then watch as it progressively got worse with more optimisations being enabled in clang. However we've since stripped the "building" portions of Dexter out as they were ill placed, which makes this tooling redundant. The lack of __init__.py adjacent to it means it couldn't be run anyway.
The core idea behind this is still sound; just it's best placed on the other side of the build system, something that Dexter shouldn't try to solve.
>From f29285e73718e379f9a91433daee91e0189bb3ff Mon Sep 17 00:00:00 2001
From: Jeremy Morse <jeremy.morse at sony.com>
Date: Fri, 21 Feb 2025 11:18:26 +0000
Subject: [PATCH] [Dexter] Remove clang-opt-bisect analyzer tool
The idea behind this tool was that you'd instrument a source file to
measure its debug-info quality, then watch as it progressively got worse
with more optimisations being enabled in clang. However we've since
stripped the "building" portions of Dexter out as they were ill placed,
which makes this tooling redundant. The lack of __init__.py adjacent to it
means it couldn't be run anyway.
The core idea behind this is still sound; just it's best placed on the
other side of the build system, something that Dexter shouldn't try to
solve.
---
.../dexter/dex/tools/clang_opt_bisect/Tool.py | 343 ------------------
1 file changed, 343 deletions(-)
delete mode 100644 cross-project-tests/debuginfo-tests/dexter/dex/tools/clang_opt_bisect/Tool.py
diff --git a/cross-project-tests/debuginfo-tests/dexter/dex/tools/clang_opt_bisect/Tool.py b/cross-project-tests/debuginfo-tests/dexter/dex/tools/clang_opt_bisect/Tool.py
deleted file mode 100644
index e6fdecb47eaae..0000000000000
--- a/cross-project-tests/debuginfo-tests/dexter/dex/tools/clang_opt_bisect/Tool.py
+++ /dev/null
@@ -1,343 +0,0 @@
-# DExTer : Debugging Experience Tester
-# ~~~~~~ ~ ~~ ~ ~~
-#
-# Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
-# See https://llvm.org/LICENSE.txt for license information.
-# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-"""Clang opt-bisect tool."""
-
-from collections import defaultdict
-import os
-import csv
-import re
-import pickle
-
-from dex.command.ParseCommand import get_command_infos
-from dex.debugger.Debuggers import run_debugger_subprocess
-from dex.debugger.DebuggerControllers.DefaultController import DefaultController
-from dex.dextIR.DextIR import DextIR
-from dex.heuristic import Heuristic
-from dex.tools import TestToolBase
-from dex.utils.Exceptions import DebuggerException, Error
-from dex.utils.Exceptions import BuildScriptException, HeuristicException
-from dex.utils.PrettyOutputBase import Stream
-from dex.utils.ReturnCode import ReturnCode
-
-
-class BisectPass(object):
- def __init__(self, no, description, description_no_loc):
- self.no = no
- self.description = description
- self.description_no_loc = description_no_loc
-
- self.penalty = 0
- self.differences = []
-
-
-class Tool(TestToolBase):
- """Use the LLVM "-opt-bisect-limit=<n>" flag to get information on the
- contribution of each LLVM pass to the overall DExTer score when using
- clang.
-
- Clang is run multiple times, with an increasing value of n, measuring the
- debugging experience at each value.
- """
-
- _re_running_pass = re.compile(r"^BISECT\: running pass \((\d+)\) (.+?)( \(.+\))?$")
-
- def __init__(self, *args, **kwargs):
- super(Tool, self).__init__(*args, **kwargs)
- self._all_bisect_pass_summary = defaultdict(list)
-
- @property
- def name(self):
- return "DExTer clang opt bisect"
-
- def _get_bisect_limits(self):
- options = self.context.options
-
- max_limit = 999999
- limits = [max_limit for _ in options.source_files]
- all_passes = [
- l
- for l in self._clang_opt_bisect_build(limits)[1].splitlines()
- if l.startswith("BISECT: running pass (")
- ]
-
- results = []
- for i, pass_ in enumerate(all_passes[1:]):
- if pass_.startswith("BISECT: running pass (1)"):
- results.append(all_passes[i])
- results.append(all_passes[-1])
-
- assert len(results) == len(options.source_files), (
- results,
- options.source_files,
- )
-
- limits = [int(Tool._re_running_pass.match(r).group(1)) for r in results]
-
- return limits
-
- def handle_options(self, defaults):
- options = self.context.options
- if "clang" not in options.builder.lower():
- raise Error(
- "--builder %s is not supported by the clang-opt-bisect tool - only 'clang' is "
- "supported " % options.builder
- )
- super(Tool, self).handle_options(defaults)
-
- def _init_debugger_controller(self):
- step_collection = DextIR(
- executable_path=self.context.options.executable,
- source_paths=self.context.options.source_files,
- dexter_version=self.context.version,
- )
-
- step_collection.commands, new_source_files = get_command_infos(
- self.context.options.source_files, self.context.options.source_root_dir
- )
- self.context.options.source_files.extend(list(new_source_files))
-
- debugger_controller = DefaultController(self.context, step_collection)
- return debugger_controller
-
- def _run_test(self, test_name): # noqa
- options = self.context.options
-
- per_pass_score = []
- current_bisect_pass_summary = defaultdict(list)
-
- max_limits = self._get_bisect_limits()
- overall_limit = sum(max_limits)
- prev_score = 1.0
- prev_steps_str = None
-
- for current_limit in range(overall_limit + 1):
- # Take the overall limit number and split it across buckets for
- # each source file.
- limit_remaining = current_limit
- file_limits = [0] * len(max_limits)
- for i, max_limit in enumerate(max_limits):
- if limit_remaining < max_limit:
- file_limits[i] += limit_remaining
- break
- else:
- file_limits[i] = max_limit
- limit_remaining -= file_limits[i]
-
- f = [l for l in file_limits if l]
- current_file_index = len(f) - 1 if f else 0
-
- _, err, builderIR = self._clang_opt_bisect_build(file_limits)
- err_lines = err.splitlines()
- # Find the last line that specified a running pass.
- for l in err_lines[::-1]:
- match = Tool._re_running_pass.match(l)
- if match:
- pass_info = match.groups()
- break
- else:
- pass_info = (0, None, None)
-
- try:
- debugger_controller = self._init_debugger_controller()
- debugger_controller = run_debugger_subprocess(
- debugger_controller, self.context.working_directory.path
- )
- steps = debugger_controller.step_collection
- except DebuggerException:
- steps = DextIR(
- executable_path=self.context.options.executable,
- source_paths=self.context.options.source_files,
- dexter_version=self.context.version,
- )
-
- steps.builder = builderIR
-
- try:
- heuristic = Heuristic(self.context, steps)
- except HeuristicException as e:
- raise Error(e)
-
- score_difference = heuristic.score - prev_score
- prev_score = heuristic.score
-
- isnan = heuristic.score != heuristic.score
- if isnan or score_difference < 0:
- color1 = "r"
- color2 = "r"
- elif score_difference > 0:
- color1 = "g"
- color2 = "g"
- else:
- color1 = "y"
- color2 = "d"
-
- summary = '<{}>running pass {}/{} on "{}"'.format(
- color2, pass_info[0], max_limits[current_file_index], test_name
- )
- if len(options.source_files) > 1:
- summary += " [{}/{}]".format(current_limit, overall_limit)
-
- pass_text = "".join(p for p in pass_info[1:] if p)
- summary += ": {} <{}>{:+.4f}</> <{}>{}</></>\n".format(
- heuristic.summary_string, color1, score_difference, color2, pass_text
- )
-
- self.context.o.auto(summary)
-
- heuristic_verbose_output = heuristic.verbose_output
-
- if options.verbose:
- self.context.o.auto(heuristic_verbose_output)
-
- steps_str = str(steps)
- steps_changed = steps_str != prev_steps_str
- prev_steps_str = steps_str
-
- # If a results directory has been specified and this is the first
- # pass or something has changed, write a text file containing
- # verbose information on the current status.
- if options.results_directory and (
- current_limit == 0 or score_difference or steps_changed
- ):
- file_name = "-".join(
- str(s)
- for s in [
- "status",
- test_name,
- "{{:0>{}}}".format(len(str(overall_limit))).format(
- current_limit
- ),
- "{:.4f}".format(heuristic.score).replace(".", "_"),
- pass_info[1],
- ]
- if s is not None
- )
-
- file_name = (
- "".join(c for c in file_name if c.isalnum() or c in "()-_./ ")
- .strip()
- .replace(" ", "_")
- .replace("/", "_")
- )
-
- output_text_path = os.path.join(
- options.results_directory, "{}.txt".format(file_name)
- )
- with open(output_text_path, "w") as fp:
- self.context.o.auto(summary + "\n", stream=Stream(fp))
- self.context.o.auto(str(steps) + "\n", stream=Stream(fp))
- self.context.o.auto(
- heuristic_verbose_output + "\n", stream=Stream(fp)
- )
-
- output_dextIR_path = os.path.join(
- options.results_directory, "{}.dextIR".format(file_name)
- )
- with open(output_dextIR_path, "wb") as fp:
- pickle.dump(steps, fp, protocol=pickle.HIGHEST_PROTOCOL)
-
- per_pass_score.append((test_name, pass_text, heuristic.score))
-
- if pass_info[1]:
- self._all_bisect_pass_summary[pass_info[1]].append(score_difference)
-
- current_bisect_pass_summary[pass_info[1]].append(score_difference)
-
- if options.results_directory:
- per_pass_score_path = os.path.join(
- options.results_directory, "{}-per_pass_score.csv".format(test_name)
- )
-
- with open(per_pass_score_path, mode="w", newline="") as fp:
- writer = csv.writer(fp, delimiter=",")
- writer.writerow(["Source File", "Pass", "Score"])
-
- for path, pass_, score in per_pass_score:
- writer.writerow([path, pass_, score])
- self.context.o.blue('wrote "{}"\n'.format(per_pass_score_path))
-
- pass_summary_path = os.path.join(
- options.results_directory, "{}-pass-summary.csv".format(test_name)
- )
-
- self._write_pass_summary(pass_summary_path, current_bisect_pass_summary)
-
- def _handle_results(self) -> ReturnCode:
- options = self.context.options
- if options.results_directory:
- pass_summary_path = os.path.join(
- options.results_directory, "overall-pass-summary.csv"
- )
-
- self._write_pass_summary(pass_summary_path, self._all_bisect_pass_summary)
- return ReturnCode.OK
-
- def _clang_opt_bisect_build(self, opt_bisect_limits):
- options = self.context.options
- compiler_options = [
- "{} -mllvm -opt-bisect-limit={}".format(options.cflags, opt_bisect_limit)
- for opt_bisect_limit in opt_bisect_limits
- ]
- linker_options = options.ldflags
-
- try:
- return run_external_build_script(
- self.context,
- source_files=options.source_files,
- compiler_options=compiler_options,
- linker_options=linker_options,
- script_path=self.build_script,
- executable_file=options.executable,
- )
- except BuildScriptException as e:
- raise Error(e)
-
- def _write_pass_summary(self, path, pass_summary):
- # Get a list of tuples.
- pass_summary_list = list(pass_summary.items())
-
- for i, item in enumerate(pass_summary_list):
- # Add elems for the sum, min, and max of the values, as well as
- # 'interestingness' which is whether any of these values are
- # non-zero.
- pass_summary_list[i] += (
- sum(item[1]),
- min(item[1]),
- max(item[1]),
- any(item[1]),
- )
-
- # Split the pass name into the basic name and kind.
- pass_summary_list[i] += tuple(item[0].rsplit(" on ", 1))
-
- # Sort the list by the following columns in order of precedence:
- # - Is interesting (True first)
- # - Sum (smallest first)
- # - Number of times pass ran (largest first)
- # - Kind (alphabetically)
- # - Name (alphabetically)
- pass_summary_list.sort(
- key=lambda tup: (not tup[5], tup[2], -len(tup[1]), tup[7], tup[6])
- )
-
- with open(path, mode="w", newline="") as fp:
- writer = csv.writer(fp, delimiter=",")
- writer.writerow(["Pass", "Kind", "Sum", "Min", "Max", "Interesting"])
-
- for (
- _,
- vals,
- sum_,
- min_,
- max_,
- interesting,
- name,
- kind,
- ) in pass_summary_list:
- writer.writerow([name, kind, sum_, min_, max_, interesting] + vals)
-
- self.context.o.blue('wrote "{}"\n'.format(path))
More information about the llvm-commits
mailing list