path: root/toolkit/crashreporter/tools/symbolstore.py

#!/bin/env python
# This Source Code Form is subject to the terms of the Mozilla Public
# License, v. 2.0. If a copy of the MPL was not distributed with this
# file, You can obtain one at http://mozilla.org/MPL/2.0/.
#
# Usage: symbolstore.py <params> <dump_syms path> <symbol store path>
#                                <debug info files or dirs>
#   Runs dump_syms on each debug info file specified on the command line,
#   then places the resulting symbol file in the proper directory
#   structure in the symbol store path.  Accepts multiple files
#   on the command line, so can be called as part of a pipe using
#   find <dir> | xargs symbolstore.pl <dump_syms> <storepath>
#   But really, you might just want to pass it <dir>.
#
#   Parameters accepted:
#     -c           : Copy debug info files to the same directory structure
#                    as sym files. On Windows, this will also copy
#                    binaries into the symbol store.
#     -a "<archs>" : Run dump_syms -a <arch> for each space separated
#                    cpu architecture in <archs> (only on OS X)
#     -s <srcdir>  : Use <srcdir> as the top source directory to
#                    generate relative filenames.

import ctypes
import errno
import os
import platform
import re
import shutil
import subprocess
import sys
import textwrap
import time
from optparse import OptionParser
from pathlib import Path

import buildconfig
from mozbuild.generated_sources import (
    GENERATED_SOURCE_EXTS,
    get_filename_with_digest,
    get_s3_region_and_bucket,
)
from mozbuild.util import memoize
from mozpack import executables
from mozpack.copier import FileRegistry
from mozpack.manifests import InstallManifest, UnreadableInstallManifest

# Utility classes


class VCSFileInfo:
    """A base class for version-controlled file information. Ensures that the
    following attributes are generated only once (successfully):

        self.root
        self.clean_root
        self.revision
        self.filename

    The attributes are generated by a single call to the GetRoot,
    GetRevision, and GetFilename methods. Those methods are explicitly not
    implemented here and must be implemented in derived classes."""

    def __init__(self, file):
        if not file:
            raise ValueError
        self.file = file

    def __getattr__(self, name):
        """__getattr__ is only called for attributes that are not set on self,
        so setting self.[attr] will prevent future calls to the GetRoot,
        GetRevision, and GetFilename methods. We don't set the values on
        failure on the off chance that a future call might succeed."""

        if name == "root":
            root = self.GetRoot()
            if root:
                self.root = root
            return root

        elif name == "clean_root":
            clean_root = self.GetCleanRoot()
            if clean_root:
                self.clean_root = clean_root
            return clean_root

        elif name == "revision":
            revision = self.GetRevision()
            if revision:
                self.revision = revision
            return revision

        elif name == "filename":
            filename = self.GetFilename()
            if filename:
                self.filename = filename
            return filename

        raise AttributeError

    def GetRoot(self):
        """This method should return the unmodified root for the file or 'None'
        on failure."""
        raise NotImplementedError

    def GetCleanRoot(self):
        """This method should return the repository root for the file or 'None'
        on failure."""
        raise NotImplementedError

    def GetRevision(self):
        """This method should return the revision number for the file or 'None'
        on failure."""
        raise NotImplementedError

    def GetFilename(self):
        """This method should return the repository-specific filename for the
        file or 'None' on failure."""
        raise NotImplementedError


# This regex separates protocol and optional username/password from a url.
# For instance, all the following urls will be transformed into
# 'foo.com/bar':
#
#   http://foo.com/bar
#   svn+ssh://user@foo.com/bar
#   svn+ssh://user:pass@foo.com/bar
#
rootRegex = re.compile(r"^\S+?:/+(?:[^\s/]*@)?(\S+)$")


def read_output(*args):
    (stdout, _) = subprocess.Popen(
        args=args, universal_newlines=True, stdout=subprocess.PIPE
    ).communicate()
    return stdout.rstrip()


class HGRepoInfo:
    def __init__(self, path):
        self.path = path

        rev = os.environ.get("MOZ_SOURCE_CHANGESET")
        if not rev:
            rev = read_output("hg", "-R", path, "parent", "--template={node}")

        # Look for the default hg path. If MOZ_SOURCE_REPO is set, we
        # don't bother asking hg.
        hg_root = os.environ.get("MOZ_SOURCE_REPO")
        if hg_root:
            root = hg_root
        else:
            root = read_output("hg", "-R", path, "showconfig", "paths.default")
            if not root:
                print("Failed to get HG Repo for %s" % path, file=sys.stderr)
        cleanroot = None
        if root:
            match = rootRegex.match(root)
            if match:
                cleanroot = match.group(1)
                if cleanroot.endswith("/"):
                    cleanroot = cleanroot[:-1]
        if cleanroot is None:
            print(
                textwrap.dedent(
                    """\
            Could not determine repo info for %s.  This is either not a clone of the web-based
            repository, or you have not specified MOZ_SOURCE_REPO, or the clone is corrupt."""
                )
                % path,
                sys.stderr,
            )
            sys.exit(1)
        self.rev = rev
        self.root = root
        self.cleanroot = cleanroot

    def GetFileInfo(self, file):
        return HGFileInfo(file, self)


class HGFileInfo(VCSFileInfo):
    def __init__(self, file, repo):
        VCSFileInfo.__init__(self, file)
        self.repo = repo
        self.file = os.path.relpath(file, repo.path)

    def GetRoot(self):
        return self.repo.root

    def GetCleanRoot(self):
        return self.repo.cleanroot

    def GetRevision(self):
        return self.repo.rev

    def GetFilename(self):
        if self.revision and self.clean_root:
            return "hg:%s:%s:%s" % (self.clean_root, self.file, self.revision)
        return self.file


class GitRepoInfo:
    """
    Info about a local git repository. Does not currently
    support discovering info about a git clone, the info must be
    provided out-of-band.
    """

    def __init__(self, path, rev, root):
        self.path = path
        cleanroot = None
        if root:
            match = rootRegex.match(root)
            if match:
                cleanroot = match.group(1)
                if cleanroot.endswith("/"):
                    cleanroot = cleanroot[:-1]
        if cleanroot is None:
            print(
                textwrap.dedent(
                    """\
            Could not determine repo info for %s (%s).  This is either not a clone of a web-based
            repository, or you have not specified MOZ_SOURCE_REPO, or the clone is corrupt."""
                )
                % (path, root),
                file=sys.stderr,
            )
            sys.exit(1)
        self.rev = rev
        self.cleanroot = cleanroot

    def GetFileInfo(self, file):
        return GitFileInfo(file, self)


class GitFileInfo(VCSFileInfo):
    def __init__(self, file, repo):
        VCSFileInfo.__init__(self, file)
        self.repo = repo
        self.file = os.path.relpath(file, repo.path)

    def GetRoot(self):
        return self.repo.path

    def GetCleanRoot(self):
        return self.repo.cleanroot

    def GetRevision(self):
        return self.repo.rev

    def GetFilename(self):
        if self.revision and self.clean_root:
            return "git:%s:%s:%s" % (self.clean_root, self.file, self.revision)
        return self.file


# Utility functions


# A cache of files for which VCS info has already been determined. Used to
# prevent extra filesystem activity or process launching.
vcsFileInfoCache = {}

if platform.system() == "Windows":

    def realpath(path):
        """
        Normalize a path using `GetFinalPathNameByHandleW` to get the
        path with all components in the case they exist in on-disk, so
        that making links to a case-sensitive server (hg.mozilla.org) works.

        This function also resolves any symlinks in the path.
        """
        # Return the original path if something fails, which can happen for paths that
        # don't exist on this system (like paths from the CRT).
        result = path

        ctypes.windll.kernel32.SetErrorMode(ctypes.c_uint(1))
        handle = ctypes.windll.kernel32.CreateFileW(
            path,
            # GENERIC_READ
            0x80000000,
            # FILE_SHARE_READ
            1,
            None,
            # OPEN_EXISTING
            3,
            # FILE_FLAG_BACKUP_SEMANTICS
            # This is necessary to open
            # directory handles.
            0x02000000,
            None,
        )
        if handle != -1:
            size = ctypes.windll.kernel32.GetFinalPathNameByHandleW(handle, None, 0, 0)
            buf = ctypes.create_unicode_buffer(size)
            if (
                ctypes.windll.kernel32.GetFinalPathNameByHandleW(handle, buf, size, 0)
                > 0
            ):
                # The return value of GetFinalPathNameByHandleW uses the
                # '\\?\' prefix.
                result = buf.value[4:]
            ctypes.windll.kernel32.CloseHandle(handle)
        return result


else:
    # Just use the os.path version otherwise.
    realpath = os.path.realpath


def IsInDir(file, dir):
    try:
        Path(file).relative_to(dir)
        return True
    except ValueError:
        return False


def GetVCSFilenameFromSrcdir(file, srcdir):
    if srcdir not in Dumper.srcdirRepoInfo:
        # Not in cache, so find it adnd cache it
        if os.path.isdir(os.path.join(srcdir, ".hg")):
            Dumper.srcdirRepoInfo[srcdir] = HGRepoInfo(srcdir)
        else:
            # Unknown VCS or file is not in a repo.
            return None
    return Dumper.srcdirRepoInfo[srcdir].GetFileInfo(file)


def GetVCSFilename(file, srcdirs):
    """Given a full path to a file, and the top source directory,
    look for version control information about this file, and return
    a tuple containing
    1) a specially formatted filename that contains the VCS type,
    VCS location, relative filename, and revision number, formatted like:
    vcs:vcs location:filename:revision
    For example:
    cvs:cvs.mozilla.org/cvsroot:mozilla/browser/app/nsBrowserApp.cpp:1.36
    2) the unmodified root information if it exists"""
    (path, filename) = os.path.split(file)
    if path == "" or filename == "":
        return (file, None)

    fileInfo = None
    root = ""
    if file in vcsFileInfoCache:
        # Already cached this info, use it.
        fileInfo = vcsFileInfoCache[file]
    else:
        for srcdir in srcdirs:
            if not IsInDir(file, srcdir):
                continue
            fileInfo = GetVCSFilenameFromSrcdir(file, srcdir)
            if fileInfo:
                vcsFileInfoCache[file] = fileInfo
                break

    if fileInfo:
        file = fileInfo.filename
        root = fileInfo.root

    # we want forward slashes on win32 paths
    return (file.replace("\\", "/"), root)


def validate_install_manifests(install_manifest_args):
    args = []
    for arg in install_manifest_args:
        bits = arg.split(",")
        if len(bits) != 2:
            raise ValueError(
                "Invalid format for --install-manifest: " "specify manifest,target_dir"
            )
        manifest_file, destination = [os.path.abspath(b) for b in bits]
        if not os.path.isfile(manifest_file):
            raise IOError(errno.ENOENT, "Manifest file not found", manifest_file)
        if not os.path.isdir(destination):
            raise IOError(errno.ENOENT, "Install directory not found", destination)
        try:
            manifest = InstallManifest(manifest_file)
        except UnreadableInstallManifest:
            raise IOError(errno.EINVAL, "Error parsing manifest file", manifest_file)
        args.append((manifest, destination))
    return args


def make_file_mapping(install_manifests):
    file_mapping = {}
    for manifest, destination in install_manifests:
        destination = os.path.abspath(destination)
        reg = FileRegistry()
        manifest.populate_registry(reg)
        for dst, src in reg:
            if hasattr(src, "path"):
                # Any paths that get compared to source file names need to go through realpath.
                abs_dest = realpath(os.path.join(destination, dst))
                file_mapping[abs_dest] = realpath(src.path)
    return file_mapping


@memoize
def get_generated_file_s3_path(filename, rel_path, bucket):
    """Given a filename, return a path formatted similarly to
    GetVCSFilename but representing a file available in an s3 bucket."""
    with open(filename, "rb") as f:
        path = get_filename_with_digest(rel_path, f.read())
        return "s3:{bucket}:{path}:".format(bucket=bucket, path=path)


def GetPlatformSpecificDumper(**kwargs):
    """This function simply returns a instance of a subclass of Dumper
    that is appropriate for the current platform."""
    return {"WINNT": Dumper_Win32, "Linux": Dumper_Linux, "Darwin": Dumper_Mac}[
        buildconfig.substs["OS_ARCH"]
    ](**kwargs)


def SourceIndex(fileStream, outputPath, vcs_root, s3_bucket):
    """Takes a list of files, writes info to a data block in a .stream file"""
    # Creates a .pdb.stream file in the mozilla\objdir to be used for source indexing
    # Create the srcsrv data block that indexes the pdb file
    result = True
    pdbStreamFile = open(outputPath, "w")
    pdbStreamFile.write(
        "SRCSRV: ini ------------------------------------------------\r\n"
        + "VERSION=2\r\n"
        + "INDEXVERSION=2\r\n"
        + "VERCTRL=http\r\n"
        + "SRCSRV: variables ------------------------------------------\r\n"
        + "SRCSRVVERCTRL=http\r\n"
        + "RUST_GITHUB_TARGET=https://github.com/rust-lang/rust/raw/%var4%/%var3%\r\n"
    )
    pdbStreamFile.write("HGSERVER=" + vcs_root + "\r\n")
    pdbStreamFile.write("HG_TARGET=%hgserver%/raw-file/%var4%/%var3%\r\n")

    if s3_bucket:
        pdbStreamFile.write("S3_BUCKET=" + s3_bucket + "\r\n")
        pdbStreamFile.write("S3_TARGET=https://%s3_bucket%.s3.amazonaws.com/%var3%\r\n")

    # Allow each entry to choose its template via "var2".
    # Possible values for var2 are: HG_TARGET / S3_TARGET / RUST_GITHUB_TARGET
    pdbStreamFile.write("SRCSRVTRG=%fnvar%(%var2%)\r\n")

    pdbStreamFile.write(
        "SRCSRV: source files ---------------------------------------\r\n"
    )
    pdbStreamFile.write(fileStream)
    pdbStreamFile.write(
        "SRCSRV: end ------------------------------------------------\r\n\n"
    )
    pdbStreamFile.close()
    return result


class Dumper:
    """This class can dump symbols from a file with debug info, and
    store the output in a directory structure that is valid for use as
    a Breakpad symbol server.  Requires a path to a dump_syms binary--
    |dump_syms| and a directory to store symbols in--|symbol_path|.
    Optionally takes a list of processor architectures to process from
    each debug file--|archs|, the full path to the top source
    directory--|srcdir|, for generating relative source file names,
    and an option to copy debug info files alongside the dumped
    symbol files--|copy_debug|, mostly useful for creating a
    Microsoft Symbol Server from the resulting output.

    You don't want to use this directly if you intend to process files.
    Instead, call GetPlatformSpecificDumper to get an instance of a
    subclass."""

    srcdirRepoInfo = {}

    def __init__(
        self,
        dump_syms,
        symbol_path,
        archs=None,
        srcdirs=[],
        copy_debug=False,
        vcsinfo=False,
        srcsrv=False,
        s3_bucket=None,
        file_mapping=None,
    ):
        # popen likes absolute paths, at least on windows
        self.dump_syms = os.path.abspath(dump_syms)
        self.symbol_path = symbol_path
        if archs is None:
            # makes the loop logic simpler
            self.archs = [""]
        else:
            self.archs = ["-a %s" % a for a in archs.split()]
        # Any paths that get compared to source file names need to go through realpath.
        self.srcdirs = [realpath(s) for s in srcdirs]
        self.copy_debug = copy_debug
        self.vcsinfo = vcsinfo
        self.srcsrv = srcsrv
        self.s3_bucket = s3_bucket
        self.file_mapping = file_mapping or {}
        # Add a static mapping for Rust sources. Since Rust 1.30 official Rust builds map
        # source paths to start with "/rust/<sha>/".
        rust_sha = buildconfig.substs["RUSTC_COMMIT"]
        rust_srcdir = "/rustc/" + rust_sha
        self.srcdirs.append(rust_srcdir)
        Dumper.srcdirRepoInfo[rust_srcdir] = GitRepoInfo(
            rust_srcdir, rust_sha, "https://github.com/rust-lang/rust/"
        )

    # subclasses override this
    def ShouldProcess(self, file):
        return True

    # This is a no-op except on Win32
    def SourceServerIndexing(
        self, debug_file, guid, sourceFileStream, vcs_root, s3_bucket
    ):
        return ""

    # subclasses override this if they want to support this
    def CopyExeAndDebugInfo(self, file, debug_file, guid, code_file, code_id):
        """This function will copy a library or executable and the file holding the
        debug information to |symbol_path|"""
        pass

    def Process(self, file_to_process, count_ctors=False):
        """Process the given file."""
        if self.ShouldProcess(os.path.abspath(file_to_process)):
            self.ProcessFile(file_to_process, count_ctors=count_ctors)

    def ProcessFile(self, file, dsymbundle=None, count_ctors=False):
        """Dump symbols from these files into a symbol file, stored
        in the proper directory structure in  |symbol_path|; processing is performed
        asynchronously, and Finish must be called to wait for it complete and cleanup.
        All files after the first are fallbacks in case the first file does not process
        successfully; if it does, no other files will be touched."""
        print("Beginning work for file: %s" % file, file=sys.stderr)

        # tries to get the vcs root from the .mozconfig first - if it's not set
        # the tinderbox vcs path will be assigned further down
        vcs_root = os.environ.get("MOZ_SOURCE_REPO")
        for arch_num, arch in enumerate(self.archs):
            self.ProcessFileWork(
                file, arch_num, arch, vcs_root, dsymbundle, count_ctors=count_ctors
            )

    def dump_syms_cmdline(self, file, arch, dsymbundle=None):
        """
        Get the commandline used to invoke dump_syms.
        """
        # The Mac dumper overrides this.
        return [self.dump_syms, "--inlines", file]

    def ProcessFileWork(
        self, file, arch_num, arch, vcs_root, dsymbundle=None, count_ctors=False
    ):
        ctors = 0
        t_start = time.time()
        print("Processing file: %s" % file, file=sys.stderr)

        sourceFileStream = ""
        code_id, code_file = None, None
        try:
            cmd = self.dump_syms_cmdline(file, arch, dsymbundle=dsymbundle)
            print(" ".join(cmd), file=sys.stderr)
            proc = subprocess.Popen(
                cmd,
                universal_newlines=True,
                stdout=subprocess.PIPE,
            )
            try:
                module_line = next(proc.stdout)
            except StopIteration:
                module_line = ""
            if module_line.startswith("MODULE"):
                # MODULE os cpu guid debug_file
                (guid, debug_file) = (module_line.split())[3:5]
                # strip off .pdb extensions, and append .sym
                sym_file = re.sub("\.pdb$", "", debug_file) + ".sym"
                # we do want forward slashes here
                rel_path = os.path.join(debug_file, guid, sym_file).replace("\\", "/")
                full_path = os.path.normpath(os.path.join(self.symbol_path, rel_path))
                try:
                    os.makedirs(os.path.dirname(full_path))
                except OSError:  # already exists
                    pass
                f = open(full_path, "w")
                f.write(module_line)
                # now process the rest of the output
                for line in proc.stdout:
                    if line.startswith("FILE"):
                        # FILE index filename
                        (x, index, filename) = line.rstrip().split(None, 2)
                        # We want original file paths for the source server.
                        sourcepath = filename
                        filename = realpath(filename)
                        if filename in self.file_mapping:
                            filename = self.file_mapping[filename]
                        if self.vcsinfo:
                            try:
                                gen_path = Path(filename)
                                rel_gen_path = gen_path.relative_to(
                                    buildconfig.topobjdir
                                )
                            except ValueError:
                                gen_path = None
                            if (
                                gen_path
                                and gen_path.exists()
                                and gen_path.suffix in GENERATED_SOURCE_EXTS
                                and self.s3_bucket
                            ):
                                filename = get_generated_file_s3_path(
                                    filename, str(rel_gen_path), self.s3_bucket
                                )
                                rootname = ""
                            else:
                                (filename, rootname) = GetVCSFilename(
                                    filename, self.srcdirs
                                )
                            # sets vcs_root in case the loop through files were to end
                            # on an empty rootname
                            if vcs_root is None:
                                if rootname:
                                    vcs_root = rootname
                        # Emit an entry for the file mapping for the srcsrv stream
                        if filename.startswith("hg:"):
                            (vcs, repo, source_file, revision) = filename.split(":", 3)
                            sourceFileStream += sourcepath + "*HG_TARGET*" + source_file
                            sourceFileStream += "*" + revision + "\r\n"
                        elif filename.startswith("s3:"):
                            (vcs, bucket, source_file, nothing) = filename.split(":", 3)
                            sourceFileStream += sourcepath + "*S3_TARGET*"
                            sourceFileStream += source_file + "\r\n"
                        elif filename.startswith("git:github.com/rust-lang/rust:"):
                            (vcs, repo, source_file, revision) = filename.split(":", 3)
                            sourceFileStream += sourcepath + "*RUST_GITHUB_TARGET*"
                            sourceFileStream += source_file + "*" + revision + "\r\n"
                        f.write("FILE %s %s\n" % (index, filename))
                    elif line.startswith("INFO CODE_ID "):
                        # INFO CODE_ID code_id code_file
                        # This gives some info we can use to
                        # store binaries in the symbol store.
                        bits = line.rstrip().split(None, 3)
                        if len(bits) == 4:
                            code_id, code_file = bits[2:]
                        f.write(line)
                    else:
                        if count_ctors and line.startswith("FUNC "):
                            # Static initializers, as created by clang and gcc
                            # have symbols that start with "_GLOBAL_sub"
                            if "_GLOBAL__sub_" in line:
                                ctors += 1
                            # MSVC creates `dynamic initializer for '...'`
                            # symbols.
                            elif "`dynamic initializer for '" in line:
                                ctors += 1

                        # pass through all other lines unchanged
                        f.write(line)
                f.close()
                retcode = proc.wait()
                if retcode != 0:
                    raise RuntimeError(
                        "dump_syms failed with error code %d while processing %s\n"
                        % (retcode, file)
                    )
                # we output relative paths so callers can get a list of what
                # was generated
                print(rel_path)
                if self.srcsrv and vcs_root:
                    # add source server indexing to the pdb file
                    self.SourceServerIndexing(
                        debug_file, guid, sourceFileStream, vcs_root, self.s3_bucket
                    )
                # only copy debug the first time if we have multiple architectures
                if self.copy_debug and arch_num == 0:
                    self.CopyExeAndDebugInfo(file, debug_file, guid, code_file, code_id)
            else:
                # For some reason, we didn't see the MODULE line as the first
                # line of output, this is strictly required so fail irrespective
                # of the process' return code.
                retcode = proc.wait()
                message = [
                    "dump_syms failed to produce the expected output",
                    "file: %s" % file,
                    "return code: %d" % retcode,
                    "first line of output: %s" % module_line,
                ]
                raise RuntimeError("\n----------\n".join(message))
        except Exception as e:
            print("Unexpected error: %s" % str(e), file=sys.stderr)
            raise

        if dsymbundle:
            shutil.rmtree(dsymbundle)

        if count_ctors:
            import json

            perfherder_data = {
                "framework": {"name": "build_metrics"},
                "suites": [
                    {
                        "name": "compiler_metrics",
                        "subtests": [
                            {
                                "name": "num_static_constructors",
                                "value": ctors,
                                "alertChangeType": "absolute",
                                "alertThreshold": 3,
                            }
                        ],
                    }
                ],
            }
            perfherder_extra_options = os.environ.get("PERFHERDER_EXTRA_OPTIONS", "")
            for opt in perfherder_extra_options.split():
                for suite in perfherder_data["suites"]:
                    if opt not in suite.get("extraOptions", []):
                        suite.setdefault("extraOptions", []).append(opt)

            if "asan" not in perfherder_extra_options.lower():
                print(
                    "PERFHERDER_DATA: %s" % json.dumps(perfherder_data), file=sys.stderr
                )

        elapsed = time.time() - t_start
        print("Finished processing %s in %.2fs" % (file, elapsed), file=sys.stderr)


# Platform-specific subclasses.  For the most part, these just have
# logic to determine what files to extract symbols from.


def locate_pdb(path):
    """Given a path to a binary, attempt to locate the matching pdb file with simple heuristics:
    * Look for a pdb file with the same base name next to the binary
    * Look for a pdb file with the same base name in the cwd

    Returns the path to the pdb file if it exists, or None if it could not be located.
    """
    path, ext = os.path.splitext(path)
    pdb = path + ".pdb"
    if os.path.isfile(pdb):
        return pdb
    # If there's no pdb next to the file, see if there's a pdb with the same root name
    # in the cwd. We build some binaries directly into dist/bin, but put the pdb files
    # in the relative objdir, which is the cwd when running this script.
    base = os.path.basename(pdb)
    pdb = os.path.join(os.getcwd(), base)
    if os.path.isfile(pdb):
        return pdb
    return None


class Dumper_Win32(Dumper):
    fixedFilenameCaseCache = {}

    def ShouldProcess(self, file):
        """This function will allow processing of exe or dll files that have pdb
        files with the same base name next to them."""
        if file.endswith(".exe") or file.endswith(".dll"):
            if locate_pdb(file) is not None:
                return True
        return False

    def CopyExeAndDebugInfo(self, file, debug_file, guid, code_file, code_id):
        """This function will copy the executable or dll and pdb files to |symbol_path|"""
        pdb_file = locate_pdb(file)

        rel_path = os.path.join(debug_file, guid, debug_file).replace("\\", "/")
        full_path = os.path.normpath(os.path.join(self.symbol_path, rel_path))
        shutil.copyfile(pdb_file, full_path)
        print(rel_path)

        # Copy the binary file as well
        if code_file and code_id:
            full_code_path = os.path.join(os.path.dirname(file), code_file)
            if os.path.exists(full_code_path):
                rel_path = os.path.join(code_file, code_id, code_file).replace(
                    "\\", "/"
                )
                full_path = os.path.normpath(os.path.join(self.symbol_path, rel_path))
                try:
                    os.makedirs(os.path.dirname(full_path))
                except OSError as e:
                    if e.errno != errno.EEXIST:
                        raise
                shutil.copyfile(full_code_path, full_path)
                print(rel_path)

    def SourceServerIndexing(
        self, debug_file, guid, sourceFileStream, vcs_root, s3_bucket
    ):
        # Creates a .pdb.stream file in the mozilla\objdir to be used for source indexing
        streamFilename = debug_file + ".stream"
        stream_output_path = os.path.abspath(streamFilename)
        # Call SourceIndex to create the .stream file
        result = SourceIndex(sourceFileStream, stream_output_path, vcs_root, s3_bucket)
        if self.copy_debug:
            pdbstr = buildconfig.substs["PDBSTR"]
            wine = buildconfig.substs.get("WINE")
            if wine:
                cmd = [wine, pdbstr]
            else:
                cmd = [pdbstr]
            subprocess.call(
                cmd
                + [
                    "-w",
                    "-p:" + os.path.basename(debug_file),
                    "-i:" + os.path.basename(streamFilename),
                    "-s:srcsrv",
                ],
                cwd=os.path.dirname(stream_output_path),
            )
            # clean up all the .stream files when done
            os.remove(stream_output_path)
        return result


class Dumper_Linux(Dumper):
    objcopy = os.environ["OBJCOPY"] if "OBJCOPY" in os.environ else "objcopy"

    def ShouldProcess(self, file):
        """This function will allow processing of files that are
        executable, or end with the .so extension, and additionally
        file(1) reports as being ELF files.  It expects to find the file
        command in PATH."""
        if file.endswith(".so") or os.access(file, os.X_OK):
            return executables.get_type(file) == executables.ELF
        return False

    def CopyExeAndDebugInfo(self, file, debug_file, guid, code_file, code_id):
        # We want to strip out the debug info, and add a
        # .gnu_debuglink section to the object, so the debugger can
        # actually load our debug info later.
        # In some odd cases, the object might already have an irrelevant
        # .gnu_debuglink section, and objcopy doesn't want to add one in
        # such cases, so we make it remove it any existing one first.
        file_dbg = file + ".dbg"
        if (
            subprocess.call([self.objcopy, "--only-keep-debug", file, file_dbg]) == 0
            and subprocess.call(
                [
                    self.objcopy,
                    "--remove-section",
                    ".gnu_debuglink",
                    "--add-gnu-debuglink=%s" % file_dbg,
                    file,
                ]
            )
            == 0
        ):
            rel_path = os.path.join(debug_file, guid, debug_file + ".dbg")
            full_path = os.path.normpath(os.path.join(self.symbol_path, rel_path))
            shutil.move(file_dbg, full_path)
            print(rel_path)
        else:
            if os.path.isfile(file_dbg):
                os.unlink(file_dbg)


class Dumper_Solaris(Dumper):
    def ShouldProcess(self, file):
        """This function will allow processing of files that are
        executable, or end with the .so extension, and additionally
        file(1) reports as being ELF files.  It expects to find the file
        command in PATH."""
        if file.endswith(".so") or os.access(file, os.X_OK):
            return executables.get_type(file) == executables.ELF
        return False


class Dumper_Mac(Dumper):
    def ShouldProcess(self, file):
        """This function will allow processing of files that are
        executable, or end with the .dylib extension, and additionally
        file(1) reports as being Mach-O files.  It expects to find the file
        command in PATH."""
        if file.endswith(".dylib") or os.access(file, os.X_OK):
            return executables.get_type(file) == executables.MACHO
        return False

    def ProcessFile(self, file, count_ctors=False):
        print("Starting Mac pre-processing on file: %s" % file, file=sys.stderr)
        dsymbundle = self.GenerateDSYM(file)
        if dsymbundle:
            # kick off new jobs per-arch with our new list of files
            Dumper.ProcessFile(
                self, file, dsymbundle=dsymbundle, count_ctors=count_ctors
            )

    def dump_syms_cmdline(self, file, arch, dsymbundle=None):
        """
        Get the commandline used to invoke dump_syms.
        """
        # dump_syms wants the path to the original binary and the .dSYM
        # in order to dump all the symbols.
        if dsymbundle:
            # This is the .dSYM bundle.
            return (
                [self.dump_syms]
                + arch.split()
                + ["--inlines", "-j", "2", dsymbundle, file]
            )
        return Dumper.dump_syms_cmdline(self, file, arch)

    def GenerateDSYM(self, file):
        """dump_syms on Mac needs to be run on a dSYM bundle produced
        by dsymutil(1), so run dsymutil here and pass the bundle name
        down to the superclass method instead."""
        t_start = time.time()
        print("Running Mac pre-processing on file: %s" % (file,), file=sys.stderr)

        dsymbundle = file + ".dSYM"
        if os.path.exists(dsymbundle):
            shutil.rmtree(dsymbundle)
        dsymutil = buildconfig.substs["DSYMUTIL"]
        # dsymutil takes --arch=foo instead of -a foo like everything else
        cmd = (
            [dsymutil] + [a.replace("-a ", "--arch=") for a in self.archs if a] + [file]
        )
        print(" ".join(cmd), file=sys.stderr)

        dsymutil_proc = subprocess.Popen(
            cmd, universal_newlines=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE
        )
        dsymout, dsymerr = dsymutil_proc.communicate()
        if dsymutil_proc.returncode != 0:
            raise RuntimeError("Error running dsymutil: %s" % dsymerr)

        # Regular dsymutil won't produce a .dSYM for files without symbols.
        if not os.path.exists(dsymbundle):
            print("No symbols found in file: %s" % (file,), file=sys.stderr)
            return False

        # llvm-dsymutil will produce a .dSYM for files without symbols or
        # debug information, but only sometimes will it warn you about this.
        # We don't want to run dump_syms on such bundles, because asserts
        # will fire in debug mode and who knows what will happen in release.
        #
        # So we check for the error message and bail if it appears.  If it
        # doesn't, we carefully check the bundled DWARF to see if dump_syms
        # will be OK with it.
        if "warning: no debug symbols in" in dsymerr:
            print(dsymerr, file=sys.stderr)
            return False

        contents_dir = os.path.join(dsymbundle, "Contents", "Resources", "DWARF")
        if not os.path.exists(contents_dir):
            print(
                "No DWARF information in .dSYM bundle %s" % (dsymbundle,),
                file=sys.stderr,
            )
            return False

        files = os.listdir(contents_dir)
        if len(files) != 1:
            print("Unexpected files in .dSYM bundle %s" % (files,), file=sys.stderr)
            return False

        otool_out = subprocess.check_output(
            [buildconfig.substs["OTOOL"], "-l", os.path.join(contents_dir, files[0])],
            universal_newlines=True,
        )
        if "sectname __debug_info" not in otool_out:
            print("No symbols in .dSYM bundle %s" % (dsymbundle,), file=sys.stderr)
            return False

        elapsed = time.time() - t_start
        print("Finished processing %s in %.2fs" % (file, elapsed), file=sys.stderr)
        return dsymbundle

    def CopyExeAndDebugInfo(self, file, debug_file, guid, code_file, code_id):
        """ProcessFile has already produced a dSYM bundle, so we should just
        copy that to the destination directory. However, we'll package it
        into a .tar because it's a bundle, so it's a directory. |file| here is
        the original filename."""
        dsymbundle = file + ".dSYM"
        rel_path = os.path.join(debug_file, guid, os.path.basename(dsymbundle) + ".tar")
        full_path = os.path.abspath(os.path.join(self.symbol_path, rel_path))
        success = subprocess.call(
            ["tar", "cf", full_path, os.path.basename(dsymbundle)],
            cwd=os.path.dirname(dsymbundle),
            stdout=open(os.devnull, "w"),
            stderr=subprocess.STDOUT,
        )
        if success == 0 and os.path.exists(full_path):
            print(rel_path)


# Entry point if called as a standalone program


def main():
    parser = OptionParser(
        usage="usage: %prog [options] <dump_syms binary> <symbol store path> <debug info files>"
    )
    parser.add_option(
        "-c",
        "--copy",
        action="store_true",
        dest="copy_debug",
        default=False,
        help="Copy debug info files into the same directory structure as symbol files",
    )
    parser.add_option(
        "-a",
        "--archs",
        action="store",
        dest="archs",
        help="Run dump_syms -a <arch> for each space separated"
        + "cpu architecture in ARCHS (only on OS X)",
    )
    parser.add_option(
        "-s",
        "--srcdir",
        action="append",
        dest="srcdir",
        default=[],
        help="Use SRCDIR to determine relative paths to source files",
    )
    parser.add_option(
        "-v",
        "--vcs-info",
        action="store_true",
        dest="vcsinfo",
        help="Try to retrieve VCS info for each FILE listed in the output",
    )
    parser.add_option(
        "-i",
        "--source-index",
        action="store_true",
        dest="srcsrv",
        default=False,
        help="Add source index information to debug files, making them suitable"
        + " for use in a source server.",
    )
    parser.add_option(
        "--install-manifest",
        action="append",
        dest="install_manifests",
        default=[],
        help="""Use this install manifest to map filenames back
to canonical locations in the source repository. Specify
<install manifest filename>,<install destination> as a comma-separated pair.""",
    )
    parser.add_option(
        "--count-ctors",
        action="store_true",
        dest="count_ctors",
        default=False,
        help="Count static initializers",
    )
    (options, args) = parser.parse_args()

    # check to see if the pdbstr.exe exists
    if options.srcsrv:
        if "PDBSTR" not in buildconfig.substs:
            print("pdbstr was not found by configure.\n", file=sys.stderr)
            sys.exit(1)

    if len(args) < 3:
        parser.error("not enough arguments")
        exit(1)

    try:
        manifests = validate_install_manifests(options.install_manifests)
    except (IOError, ValueError) as e:
        parser.error(str(e))
        exit(1)
    file_mapping = make_file_mapping(manifests)
    _, bucket = get_s3_region_and_bucket()
    dumper = GetPlatformSpecificDumper(
        dump_syms=args[0],
        symbol_path=args[1],
        copy_debug=options.copy_debug,
        archs=options.archs,
        srcdirs=options.srcdir,
        vcsinfo=options.vcsinfo,
        srcsrv=options.srcsrv,
        s3_bucket=bucket,
        file_mapping=file_mapping,
    )

    dumper.Process(args[2], options.count_ctors)


# run main if run directly
if __name__ == "__main__":
    main()