#!/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 # # 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 | xargs symbolstore.pl # But really, you might just want to pass it . # # Parameters accepted: # -c : Copy debug info files to the same directory structure # as sym files # -a "" : Run dump_syms -a for each space separated # cpu architecture in (only on OS X) # -s : Use as the top source directory to # generate relative filenames. import sys import platform import os import re import shutil import textwrap import fnmatch import subprocess import urlparse from optparse import OptionParser from xml.dom.minidom import parse # 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 NotImplementedErrors 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, stdout=subprocess.PIPE).communicate() return stdout.rstrip() class HGRepoInfo: def __init__(self, path): self.path = path rev = read_output('hg', '-R', path, 'parent', '--template={node|short}') # Look for the default hg path. If SRVSRV_ROOT is set, we # don't bother asking hg. hg_root = os.environ.get("SRCSRV_ROOT") if hg_root: root = hg_root else: root = read_output('hg', '-R', path, 'showconfig', 'paths.default') if not root: print >> sys.stderr, "Failed to get HG Repo for %s" % 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 >> sys.stderr, 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 SRCSRV_ROOT, or the clone is corrupt.""") % path sys.exit(1) self.rev = rev 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.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 "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 >> sys.stderr, 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 SRCSRV_ROOT, or the clone is corrupt.""") % (path, root) 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 repo info for each srcdir. srcdirRepoInfo = {} # A cache of files for which VCS info has already been determined. Used to # prevent extra filesystem activity or process launching. vcsFileInfoCache = {} def IsInDir(file, dir): # the lower() is to handle win32+vc8, where # the source filenames come out all lowercase, # but the srcdir can be mixed case return os.path.abspath(file).lower().startswith(os.path.abspath(dir).lower()) def GetVCSFilenameFromSrcdir(file, srcdir): if srcdir not in srcdirRepoInfo: # Not in cache, so find it adnd cache it if os.path.isdir(os.path.join(srcdir, '.hg')): srcdirRepoInfo[srcdir] = HGRepoInfo(srcdir) else: # Unknown VCS or file is not in a repo. return None return 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 GetPlatformSpecificDumper(**kwargs): """This function simply returns a instance of a subclass of Dumper that is appropriate for the current platform.""" # Python 2.5 has a bug where platform.system() returns 'Microsoft'. # Remove this when we no longer support Python 2.5. return {'Windows': Dumper_Win32, 'Microsoft': Dumper_Win32, 'Linux': Dumper_Linux, 'Sunos5': Dumper_Solaris, 'Darwin': Dumper_Mac}[platform.system()](**kwargs) def SourceIndex(fileStream, outputPath, vcs_root): """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\nVERSION=2\r\nINDEXVERSION=2\r\nVERCTRL=http\r\nSRCSRV: variables ------------------------------------------\r\nHGSERVER=''') pdbStreamFile.write(vcs_root) pdbStreamFile.write('''\r\nSRCSRVVERCTRL=http\r\nHTTP_EXTRACT_TARGET=%hgserver%/raw-file/%var3%/%var2%\r\nSRCSRVTRG=%http_extract_target%\r\nSRCSRV: source files ---------------------------------------\r\n''') pdbStreamFile.write(fileStream) # can't do string interpolation because the source server also uses this and so there are % in the above 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 call ProcessDir. Instead, call GetPlatformSpecificDumper to get an instance of a subclass.""" def __init__(self, dump_syms, symbol_path, archs=None, srcdirs=[], copy_debug=False, vcsinfo=False, srcsrv=False, exclude=[], repo_manifest=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()] self.srcdirs = [os.path.normpath(a) for a in srcdirs] self.copy_debug = copy_debug self.vcsinfo = vcsinfo self.srcsrv = srcsrv self.exclude = exclude[:] if repo_manifest: self.parse_repo_manifest(repo_manifest) def parse_repo_manifest(self, repo_manifest): """ Parse an XML manifest of repository info as produced by the `repo manifest -r` command. """ doc = parse(repo_manifest) if doc.firstChild.tagName != "manifest": return # First, get remotes. remotes = dict([(r.getAttribute("name"), r.getAttribute("fetch")) for r in doc.getElementsByTagName("remote")]) # And default remote. default_remote = None if doc.getElementsByTagName("default"): default_remote = doc.getElementsByTagName("default")[0].getAttribute("remote") # Now get projects. Assume they're relative to repo_manifest. base_dir = os.path.abspath(os.path.dirname(repo_manifest)) for proj in doc.getElementsByTagName("project"): # name is the repository URL relative to the remote path. name = proj.getAttribute("name") # path is the path on-disk, relative to the manifest file. path = proj.getAttribute("path") # revision is the changeset ID. rev = proj.getAttribute("revision") # remote is the base URL to use. remote = proj.getAttribute("remote") # remote defaults to the . if not remote: remote = default_remote # path defaults to name. if not path: path = name if not (name and path and rev and remote): print "Skipping project %s" % proj.toxml() continue remote = remotes[remote] # Turn git URLs into http URLs so that urljoin works. if remote.startswith("git:"): remote = "http" + remote[3:] # Add this project to srcdirs. srcdir = os.path.join(base_dir, path) self.srcdirs.append(srcdir) # And cache its VCS file info. Currently all repos mentioned # in a repo manifest are assumed to be git. root = urlparse.urljoin(remote, name) srcdirRepoInfo[srcdir] = GitRepoInfo(srcdir, rev, root) # subclasses override this def ShouldProcess(self, file): return not any(fnmatch.fnmatch(os.path.basename(file), exclude) for exclude in self.exclude) # and can override this def ShouldSkipDir(self, dir): return False def RunFileCommand(self, file): """Utility function, returns the output of file(1)""" try: # we use -L to read the targets of symlinks, # and -b to print just the content, not the filename return os.popen("file -Lb " + file).read() except: return "" # This is a no-op except on Win32 def FixFilenameCase(self, file): return file # This is a no-op except on Win32 def SourceServerIndexing(self, debug_file, guid, sourceFileStream, vcs_root): return "" # subclasses override this if they want to support this def CopyDebug(self, file, debug_file, guid): pass def Process(self, file_or_dir): "Process a file or all the (valid) files in a directory." if os.path.isdir(file_or_dir) and not self.ShouldSkipDir(file_or_dir): return self.ProcessDir(file_or_dir) elif os.path.isfile(file_or_dir): return self.ProcessFile(file_or_dir) # maybe it doesn't exist? return False def ProcessDir(self, dir): """Process all the valid files in this directory. Valid files are determined by calling ShouldProcess.""" result = True for root, dirs, files in os.walk(dir): for d in dirs[:]: if self.ShouldSkipDir(d): dirs.remove(d) for f in files: fullpath = os.path.join(root, f) if self.ShouldProcess(fullpath): if not self.ProcessFile(fullpath): result = False return result def ProcessFile(self, file): """Dump symbols from this file into a symbol file, stored in the proper directory structure in |symbol_path|.""" print >> sys.stderr, "Processing file: %s" % file sys.stderr.flush() result = False sourceFileStream = '' # 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("SRCSRV_ROOT") for arch_num, arch in enumerate(self.archs): try: proc = subprocess.Popen([self.dump_syms] + arch.split() + [file], stdout=subprocess.PIPE) module_line = proc.stdout.next() 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) if sys.platform == "sunos5": for srcdir in self.srcdirs: start = filename.find(self.srcdir) if start != -1: filename = filename[start:] break filename = self.FixFilenameCase(filename) sourcepath = filename if self.vcsinfo: (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 # gather up files with hg for indexing if filename.startswith("hg"): (ver, checkout, source_file, revision) = filename.split(":", 3) sourceFileStream += sourcepath + "*" + source_file + '*' + revision + "\r\n" f.write("FILE %s %s\n" % (index, filename)) else: # pass through all other lines unchanged f.write(line) # we want to return true only if at least one line is not a MODULE or FILE line result = True f.close() proc.wait() # 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(file, guid, sourceFileStream, vcs_root) # only copy debug the first time if we have multiple architectures if self.copy_debug and arch_num == 0: self.CopyDebug(file, debug_file, guid) except StopIteration: pass except: print >> sys.stderr, "Unexpected error: ", sys.exc_info()[0] raise return result # Platform-specific subclasses. For the most part, these just have # logic to determine what files to extract symbols from. class Dumper_Win32(Dumper): fixedFilenameCaseCache = {} def ShouldProcess(self, file): """This function will allow processing of pdb files that have dll or exe files with the same base name next to them.""" if not Dumper.ShouldProcess(self, file): return False if file.endswith(".pdb"): (path,ext) = os.path.splitext(file) if os.path.isfile(path + ".exe") or os.path.isfile(path + ".dll"): return True return False def FixFilenameCase(self, file): """Recent versions of Visual C++ put filenames into PDB files as all lowercase. If the file exists on the local filesystem, fix it.""" # Use a cached version if we have one. if file in self.fixedFilenameCaseCache: return self.fixedFilenameCaseCache[file] result = file (path, filename) = os.path.split(file) if os.path.isdir(path): lc_filename = filename.lower() for f in os.listdir(path): if f.lower() == lc_filename: result = os.path.join(path, f) break # Cache the corrected version to avoid future filesystem hits. self.fixedFilenameCaseCache[file] = result return result def CopyDebug(self, file, debug_file, guid): 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(file, full_path) # try compressing it compressed_file = os.path.splitext(full_path)[0] + ".pd_" # ignore makecab's output success = subprocess.call(["makecab.exe", "/D", "CompressionType=LZX", "/D", "CompressionMemory=21", full_path, compressed_file], stdout=open("NUL:","w"), stderr=subprocess.STDOUT) if success == 0 and os.path.exists(compressed_file): os.unlink(full_path) print os.path.splitext(rel_path)[0] + ".pd_" else: print rel_path def SourceServerIndexing(self, debug_file, guid, sourceFileStream, vcs_root): # Creates a .pdb.stream file in the mozilla\objdir to be used for source indexing debug_file = os.path.abspath(debug_file) 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) if self.copy_debug: pdbstr_path = os.environ.get("PDBSTR_PATH") pdbstr = os.path.normpath(pdbstr_path) subprocess.call([pdbstr, "-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 not Dumper.ShouldProcess(self, file): return False if file.endswith(".so") or os.access(file, os.X_OK): return self.RunFileCommand(file).startswith("ELF") return False def CopyDebug(self, file, debug_file, guid): # 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. file_dbg = file + ".dbg" if subprocess.call([self.objcopy, '--only-keep-debug', file, file_dbg]) == 0 and \ subprocess.call([self.objcopy, '--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) # gzip the shipped debug files os.system("gzip %s" % full_path) print rel_path + ".gz" else: if os.path.isfile(file_dbg): os.unlink(file_dbg) class Dumper_Solaris(Dumper): def RunFileCommand(self, file): """Utility function, returns the output of file(1)""" try: output = os.popen("file " + file).read() return output.split('\t')[1]; except: return "" 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 not Dumper.ShouldProcess(self, file): return False if file.endswith(".so") or os.access(file, os.X_OK): return self.RunFileCommand(file).startswith("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 not Dumper.ShouldProcess(self, file): return False if file.endswith(".dylib") or os.access(file, os.X_OK): return self.RunFileCommand(file).startswith("Mach-O") return False def ShouldSkipDir(self, dir): """We create .dSYM bundles on the fly, but if someone runs buildsymbols twice, we should skip any bundles we created previously, otherwise we'll recurse into them and try to dump the inner bits again.""" if dir.endswith(".dSYM"): return True return False def ProcessFile(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.""" dsymbundle = file + ".dSYM" if os.path.exists(dsymbundle): shutil.rmtree(dsymbundle) # dsymutil takes --arch=foo instead of -a foo like everything else subprocess.call(["dsymutil"] + [a.replace('-a ', '--arch=') for a in self.archs if a] + [file], stdout=open("/dev/null","w")) if not os.path.exists(dsymbundle): # dsymutil won't produce a .dSYM for files without symbols return False res = Dumper.ProcessFile(self, dsymbundle) # CopyDebug will already have been run from Dumper.ProcessFile shutil.rmtree(dsymbundle) # fallback for DWARF-less binaries if not res: print >> sys.stderr, "Couldn't read DWARF symbols in: %s" % dsymbundle res = Dumper.ProcessFile(self, file) return res def CopyDebug(self, file, debug_file, guid): """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.bz2 because the debug symbols are pretty huge, and also because it's a bundle, so it's a directory. |file| here is the dSYM bundle, and |debug_file| is the original filename.""" rel_path = os.path.join(debug_file, guid, os.path.basename(file) + ".tar.bz2") full_path = os.path.abspath(os.path.join(self.symbol_path, rel_path)) success = subprocess.call(["tar", "cjf", full_path, os.path.basename(file)], cwd=os.path.dirname(file), stdout=open("/dev/null","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] ") 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 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("-x", "--exclude", action="append", dest="exclude", default=[], metavar="PATTERN", help="Skip processing files matching PATTERN.") parser.add_option("--repo-manifest", action="store", dest="repo_manifest", help="""Get source information from this XML manifest produced by the `repo manifest -r` command. """) (options, args) = parser.parse_args() #check to see if the pdbstr.exe exists if options.srcsrv: pdbstr = os.environ.get("PDBSTR_PATH") if not os.path.exists(pdbstr): print >> sys.stderr, "Invalid path to pdbstr.exe - please set/check PDBSTR_PATH.\n" sys.exit(1) if len(args) < 3: parser.error("not enough arguments") exit(1) 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, exclude=options.exclude, repo_manifest=options.repo_manifest) for arg in args[2:]: dumper.Process(arg) # run main if run directly if __name__ == "__main__": main()