Miscellaneous improvements to minidump-2-core:...

Miscellaneous improvements to minidump-2-core:

 - now builds from gyp.

 - added a "-v" flag to print (almost) all of the information contained in
   the minidump file to stderr.

 - keep track of all mappings including the data that is part of this
   mapping (if any). This allows us to read the entire minidump file
   initially, combine all the different sources of information that we
   have, and come up with a complete picture of what needs to go into the
   core file. Previously, we had a hard time combining partial data from
   different minidump streams.

 - parse all information about memory mappings (including the /proc/$PID/maps
   data) that we get from the minidump. This gives us the most complete
   core file possible.

 - if breakpad recorded the loaded DSOs, update the _DYNAMIC data structure
   in the core file to include this information. The exact details for this
   data structure are described in <link.h>.

 - keep track of signatures for all DSOs. When we then generate the core
   file, include the signature in the filename of the DSO. This allows us
   to analyze user-provided crash dumps, even if the user has system
   libraries that differ from ours.
   We now assume that all of these DSOs are installed in "/var/lib/breakpad/"

 - added parsing of CPU register information for x86-64. Previously, we could
   only parse minidumps generated on i386.

 - detect incorrectly written minidump files (e.g. zero'd out environment and
   corrupted command lines) and don't bother parsing defective streams.

 - detect if the minidump file used the wrong stream identifier for the
   AUXV vector and compensate for this bug.

TEST=convert minidump files, load in gdb, check that we get full backtraces
BUG=none

[Markus (顧孟勤), 2010-08-12 04:03:24]

I made some minidump-2-core changes that give us much more complete core files
from our minidumps.

These changes really only mark sense though, once my other change list goes in.
I made changes to breakpad that make it include information about DSOs into the
minidump. This information can then be used by "gdb" to find the appropriate
DSOs and to give us correct backtraces.

[Lei Zhang, 2010-08-12 20:39:00]

breakpad/src is actually http://google-breakpad.googlecode.com/svn/trunk. You
need to upload a CL to breakpad.appspot.com and get it reviewed there. Then we
can roll DEPS in Chromium and commit the breakpad.gyp changes.

[agl, 2010-08-12 21:03:56]

LGTM

http://codereview.chromium.org/3152010/diff/1/4
File breakpad/src/tools/linux/md2core/minidump-2-core.cc (right):

http://codereview.chromium.org/3152010/diff/1/4#newcode135
breakpad/src/tools/linux/md2core/minidump-2-core.cc:135: // Get a zero-terminate
string. This method only works correctly for ASCII
typo: terminated

http://codereview.chromium.org/3152010/diff/1/4#newcode519
breakpad/src/tools/linux/md2core/minidump-2-core.cc:519: for (const u_int8_t*
ptr = range.data();
You gotta love C code with: uint8 (Chromium), uint8_t (C99) and u_int8_t here,
all trying to say 'byte' :)

http://codereview.chromium.org/3152010/diff/1/4#newcode801
breakpad/src/tools/linux/md2core/minidump-2-core.cc:801: std::string basename =
slash == std::string::npos
I think the ? should end up on the end of the previous line, but don't really
care.

http://codereview.chromium.org/3152010/diff/1/4#newcode824
breakpad/src/tools/linux/md2core/minidump-2-core.cc:824: for (std::map<uint64_t,
CrashedProcess::Mapping>::iterator iter =
Formatting needs some fixing here.

http://codereview.chromium.org/3152010/diff/1/4#newcode837
breakpad/src/tools/linux/md2core/minidump-2-core.cc:837: mapping.offset +=
iter->second.end_address -
I must admit that I don't understand why you're doing this line.

http://codereview.chromium.org/3152010/diff/1/4#newcode923
breakpad/src/tools/linux/md2core/minidump-2-core.cc:923: if
(crashinfo->dynamic_data != "") {
if (!crashinfo->dynamic_data.empty()) {