* Switch to Chromium's clang package on Windows. Fuchsia's clang
package uses symlinks, which don't Just Work. See comment 3
on the linked bug for details.
* Also tweak the installed path of the clang package to use
win-amd64 instead of windows-amd64 to match gn's host_os
variable, which is needed by the mini_chromium change in the
roll mentioned below.
* Add `__declspec(noinline)` to sources of some end-to-end test
binaries where the test expects a certain stack layout that's
perturbed by inlining
* self_destroying_test_program.cc failed with Access Violation
instead of with Breakpoint. I'm not 100% sure what's happening
there as I couldn't reproduce it on my machine. The test uses
VirtualFree(..., MEM_DECOMMIT) to make parts of the stack unreadable
to make sure crashpad can handle that. My theory is that clang
optimized out the call to `_alloca()`, which results in not enough
of the stack being around for VirtualFree() to be able to decommit.
https://geidav.wordpress.com/tag/reserved-memory/#:~:text=How%20does%20Windows%20manage%20stack%20memory%3F
and https://godbolt.org/z/YdbYdKeMh suggest that this theory is
not completely off -- but the
`[[maybe_unused]] volatile void* do_not_optimize_away` bit feels
homeopathic. And yet, with it this seems to pass CQ (see try results
at patch sets 12 and 20) and without it it doesn't (see patch set 19).
Maybe this was just luck and the test still feels flakily!
(The linker's `/STACK` defaults to 1MB stack reserved and 4K
committed -- 4K kind of seems like it should be enough for that
VirtualFree call to succeed if that really is the problem.)
So this part is a bit experimental.
Anyways, nothing uses clang yet, so no behavior change at this point
(other than `gclient sync` downloading less stuff and not failing
by default on Windows).
Includes a one-rev mini_chromium roll:
12ef786772..08d490553b
$ git log 12ef78677..08d490553 --date=short --no-merges --format='%ad %ae %s'
2025-02-13 thakis Prepare mini_chromium for using clang on Windows
Bug: 384682775
Change-Id: I33b54acf18cb6eadfd2167c76c0b4a5824efa64d
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/6242577
Reviewed-by: Mark Mentovai <mark@chromium.org>
Commit-Queue: Nico Weber <thakis@chromium.org>
__builtin_trap() is already noreturn, so no need to put a
NOTREACHED() after it.
No behavior change.
Bug: chromium:393924694
Change-Id: Ib93b429d8ef7cfe8fb59c4b9e9b8afced06b2ac2
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/6298671
Commit-Queue: Nico Weber <thakis@chromium.org>
Reviewed-by: Mark Mentovai <mark@chromium.org>
gyp is no longer used in/by crashpad (since crrev.com/c/3034984) and
isn't referenced in DEPS.
Bug: 374516744
Change-Id: I06bf6d05457a9f0a084db8ec2da75b6e07780b41
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/6168877
Commit-Queue: Peter Boström <pbos@chromium.org>
Reviewed-by: Mark Mentovai <mark@chromium.org>
Not used for anything yet.
Bug: chromium:384682775
Change-Id: I00a379756633b0edc849fa1d8bbd6e81f44a6ab1
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/6214931
Reviewed-by: Mark Mentovai <mark@chromium.org>
Commit-Queue: Nico Weber <thakis@chromium.org>
doc/support/crashpad.doxy is updated with:
```
% doxygen -u doc/support/crashpad.doxy
[…]
% doxygen -v
1.13.2
```
Additional updates to doc/support/crashpad.doxy are made:
- The settings for `DISABLE_INDEX` and `GENERATE_TREEVIEW` are updated,
as Doxygen’s defaults have changed. This switches from the
top-of-page index to the side-of-page tree view.
- The TODO and Deprecated sections have become more prominent under the
side-of-page tree view than they were with the top-of-page index, and
they’re not very useful in Crashpad documentation, so they’re
disabled by setting `GENERATE_TODOLIST` and `GENERATE_DEPRECATEDLIST`
appropriately. The similar (but unused in Crashpad)
`GENERATE_TESTLIST` and `GENERATE_BUGLIST` are also disabled.
- `USE_DOT` is now set, to use `dot` from GraphViz to generate SVG
diagrams. These look better than the PNGs that Doxygen generated
without `dot`. `DOT_COMMON_ATTR` and `DOT_EDGE_ATTR` are set to make
the fonts used in GraphViz-generated SVGs match those used in
Doxygen-generated HTML/CSS.
- `EXCLUDE` has been updated to drop directories that no longer exist
(compat/non_cxx11_lib) and add ones that now do (compat/android,
compat/ios, and compat/linux).
- Some values that were otherwise unused in doc/support/crashpad.doxy
are set back to their default values. The differences from the
default configuration are shown with `doxygen -x
doc/support/crashpad.doxy`, and a configuration template with default
values can be generated for inspection with `doxygen -g
/tmp/template.doxy`. The tags affected are:
- `MATHJAX_RELPATH`, unused since `USE_MATHJAX` is `NO`.
- `LATEX_CMD_NAME` and `LATEX_BIB_STYLE`, unused since `USE_LATEX`
is `NO`).
- `EXCLUDE_PATTERNS`, unused since `EXCLUDE_PATH` is empty.
doc/support/doxygen_crashpad.css is updated to change the non-monospaced
font from Open Sans to Noto Sans. It is also updated to use the CSS
properties Doxygen defines for the purpose, rather than monkeying around
with custom selectors. The properties can be discovered by running
`doxygen -w html header.html footer.html customdoxygen.css
doc/support/crashpad.doxy` and reviewing customdoxygen.css.
doc/support/generate.sh is updated to further execute
doc/support/generate_doxygen.py by invoking directly and relying on its
`#!/usr/bin/env python3` line rather than invoking via `python`, which
is no longer available on many systems. doc/support/generate_doxygen.py
has already been Python 3-compatible since f88a116c0e2e
(https://chromium-review.googlesource.com/c/3542572, 2022-03-23).
Finally, there are a number of changes to fix Doxygen warnings produced
by the new Doxygen version or because of new code that has not yet been
run through Doxygen during a test run.
Change-Id: I436688b16530cb0a07dbf89d32601fff689ac2f2
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/6180234
Reviewed-by: Justin Cohen <justincohen@chromium.org>
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
Commit-Queue: Mark Mentovai <mark@chromium.org>
Includes an update of mini_chromium to
12ef786772d9a73751e2d0f3ef9c792b09c386b5.
12ef786772d9 OWN mini_chromium
Bug: 390190633
Change-Id: Icfe39a08dc439125c53cb7da1ed7ef69850de924
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/6177891
Reviewed-by: Justin Cohen <justincohen@chromium.org>
Reviewed-by: Joey Scarr <jsca@google.com>
Commit-Queue: Mark Mentovai <mark@chromium.org>
When starting or restarting Chromium on Windows, we see the busy cursor
briefly. This is because crashpad does not specify
STARTF_FORCEOFFFEEDBACK in the CreateProcess call. This is similar to
other changes in Chromium for child processes to avoid the busy cursor.
Bug: 389978731
Change-Id: Ib95cd3b6ce13e3170324e63b43e46a376affc7c8
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/6173930
Reviewed-by: Will Harris <wfh@google.com>
Commit-Queue: Chris Davis <chrdavis@microsoft.com>
Reviewed-by: Peter Boström <pbos@chromium.org>
Reviewed-by: Mark Mentovai <mark@chromium.org>
Reviewed-by: Will Harris <wfh@chromium.org>
VS Code creates a .vscode/ directory in the top level of the repo
to hold config files.
Adding it to .gitignore will make it easier to develop crashpad using
VS Code.
Change-Id: I7afbbbf8ed6c017edc0299a18bf7ad0c11407b93
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/6013035
Reviewed-by: Mark Mentovai <mark@chromium.org>
Commit-Queue: Jesse McKenna <jessemckenna@google.com>
A future libc++ revision will stop providing this include transitively.
Bug: 376278210
Change-Id: I79460933ed915bc601278a6d75f2c2ced1130440
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/6049220
Commit-Queue: Nico Weber <thakis@chromium.org>
Reviewed-by: Mark Mentovai <mark@chromium.org>
- Where std::nullptr_t is used (in namespace std), #include <cstddef>,
even if it appears alongside <stddef.h> because types like intptr_t
are used (in the root namespace).
- Where math library functions are used, #include <cmath> to get and
use the versions of functions in namespace std which are overloaded
by argument type, such as std::isinf and std::isnan.
third_party is excluded.
Change-Id: Ic9c143f407e5a9c5cff4098cf566ba5caf4bf6d8
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/6021439
Commit-Queue: Mark Mentovai <mark@chromium.org>
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
In most cases NOTREACHED() is now a better option. Also performs
dead-code removal.
Bug: 40122554, 40580068
Change-Id: I2deffa59d375db7813ea6779b1d387718d5c62c7
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/6013409
Reviewed-by: Mark Mentovai <mark@chromium.org>
Commit-Queue: Peter Boström <pbos@chromium.org>
This attempts to be somewhat forward-compatible with upcoming additions
to the Data struct, most importantly to not lose the client ID if we
ever need to downgrade / read data from a future crashpad version.
Bug: 42310127
Change-Id: Ic3914fdd8460f4f41e5bb523d5c52361767880dd
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/5915193
Reviewed-by: Mark Mentovai <mark@chromium.org>
Reviewed-by: Jesse McKenna <jessemckenna@google.com>
Commit-Queue: Peter Boström <pbos@chromium.org>
Routing LogSink manually is not needed, as this is done implicitly
through the client shard. Future CLs will update the routing used in
the shard, so it's necessary to cleanup these duplicate routes in
order to update the shards.
Bug: fuchsia:345827642
Change-Id: I529ab52412530382240d5c6276c1ef01ab349ecc
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/5941579
Reviewed-by: Mark Mentovai <mark@chromium.org>
Commit-Queue: Brian Bosak <bbosak@google.com>
buildtools has first party support to fetch clang-format so no longer
needed to use custom hooks.
Bug: chromium:336843583
Change-Id: I2c29ad9df579b605ff6e1d967da6502ffcd12faf
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/5586422
Commit-Queue: Josip Sokcevic <sokcevic@chromium.org>
Reviewed-by: Mark Mentovai <mark@chromium.org>
The gn-args have default values in fuchsia-gn-sdk and not necessary
to be repeated - api-level 18 is also removed and breaks crashpad.
Bug: crashpad:365889763
Change-Id: I2a5a573158db434dd578c92455d0af6c96408d74
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/5854005
Reviewed-by: Mark Mentovai <mark@chromium.org>
Commit-Queue: Mark Mentovai <mark@chromium.org>
This GN arg config repros the failure in the bug:
```
chrome_pgo_phase = 1
is_official_build = true
target_cpu = "arm64"
target_os = "android"
use_remoteexec = true
```
Commenting out either is_official_build or chrome_pgo_phase removes the
compile error, but since many bots build with is_official_build, the
culprit must be chrome_pgo_phase.
Bug: chromium:362787700
Change-Id: I536a098693b85ed93c22e38e639b393fcb480e79
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/5826154
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
`DISABLE_CFI_ICALL` is already defined in that header; use it.
This is both simpler and less likely to trigger macro redefinition
errors.
This was previously landed in crrev.com/c/5805458, but reverted due to
crbug.com/362787700.
Bug: chromium:362787700
Change-Id: I1fc8d9dc0b099c67ea7d774034b7691691ff2a50
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/5825328
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
`DISABLE_CFI_ICALL` is already defined in that header; use it.
This is both simpler and less likely to trigger macro redefinition
errors.
Bug: none
Change-Id: I3ecfe9b6dc4ac42c6a69b3fd9c6d2c68fe8e62a2
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/5805458
Reviewed-by: Mark Mentovai <mark@chromium.org>
There is no longer a need to use this macro; compilers have no problem
with `alignas()` nowadays.
Subsequent CLs will remove the macro entirely once it's unused in both
Crashpad and Chromium.
Bug: none
Change-Id: I94675f3f674f9bc32a213e972a017980dcd1c014
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/5805982
Reviewed-by: Mark Mentovai <mark@chromium.org>
Commit-Queue: Mark Mentovai <mark@chromium.org>
This was tested locally by adding "-Wunreachable-code-aggressive" after
making NOTREACHED() [[noreturn]] in mini_chromium and then getting that
to compile.
Bug: chromium:40580068
Change-Id: I7ec1c72be1d73436d128660a621e9060eaebaee8
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/5780891
Reviewed-by: Mark Mentovai <mark@chromium.org>
This alias is not present in mini_chromium and disappearing from
chromium.
Bug: chromium:40580068
Change-Id: Ib6218d96df3bb8b2fe46af08ac51d6362c7db64e
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/5783604
Reviewed-by: Mark Mentovai <mark@chromium.org>
This includes an upstreamed version of the Crashpad-local 1a62a0182557
(https://chromium-review.googlesource.com/c/5769752) for BoringSSL
compatibility.
50fce538c685 threadsafe CLOEXEC on platforms that support it
fb739dbaecf3 threadsafe accept on windows, linux
ed0719f2bcb5 Code format
521529d24d38 Fix#1481 (with content provider) (#1527)
e00fd06355e8 Release v0.16.1
ff038f98b701 Merge branch 'thread-safe-cloexec' of
github.com:kdombroski/cpp-httplib into kdombroski-thread-
safe-cloexec
ae63b89cbf70 Use SOCK_CLOEXEC instead of __linux__
69c84c9597c3 BoringSSL compatibility fixes (#1892)
c5c54b31e2bf Release v0.16.2
Change-Id: I01485010de53ae599e22c8ce3c9d6af046d47c24
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/5769660
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
This fixes errors observed while building
util/http_transport_test_server/http_transport_test_server.cc, shown
below.
The fixes include:
- Library version check: tolerate BoringSSL as an alternative to
OpenSSL 3.
- Don’t call `OPENSSL_thread_stop`, which is not in BoringSSL.
- Use `SSL_get_peer_certificate` (deprecated in OpenSSL 3), the old
name for `SSL_get1_peer_certificate`, because the new name is not in
BoringSSL.
- Call `SSL_set_tlsext_host_name` directly instead of making a quirky
`SSL_ctrl` call that BoringSSL does not support. The feared
-Wold-style-cast warning that occurs when buidling with OpenSSL is
not triggered in BoringSSL.
Compilation errors from
https://chromium-review.googlesource.com/c/5766975?checksPatchset=1&tab=checks
→ https://ci.chromium.org/ui/p/chromium/builders/try/linux-rel/1909715/
→ “10. compilator steps (with patch)” → “31. compile (with patch)” →
stdout
(https://logs.chromium.org/logs/chromium/buildbucket/cr-buildbucket/8740323272553670737/+/u/compile__with_patch_/stdout):
```
In file included from util/net/http_transport_test_server.cc:42:
third_party/cpp-httplib/cpp-httplib/httplib.h:275:2: error: Sorry, OpenSSL versions prior to 3.0.0 are not supported
275 | #error Sorry, OpenSSL versions prior to 3.0.0 are not supported
| ^
In file included from util/net/http_transport_test_server.cc:42:
third_party/cpp-httplib/cpp-httplib/httplib.h:733:7: error: use of undeclared identifier 'OPENSSL_thread_stop'
733 | OPENSSL_thread_stop ();
| ^
third_party/cpp-httplib/cpp-httplib/httplib.h:9062:30: error: use of undeclared identifier 'SSL_get1_peer_certificate'; did you mean 'SSL_get_peer_certificate'?
9062 | auto server_cert = SSL_get1_peer_certificate(ssl2);
| ^~~~~~~~~~~~~~~~~~~~~~~~~
| SSL_get_peer_certificate
…/boringssl/src/include/openssl/ssl.h:1784:22: note: 'SSL_get_peer_certificate' declared here
1784 | OPENSSL_EXPORT X509 *SSL_get_peer_certificate(const SSL *ssl);
| ^
In file included from util/net/http_transport_test_server.cc:42:
third_party/cpp-httplib/cpp-httplib/httplib.h:9083:24: error: use of undeclared identifier 'doesnt_exist'
9083 | SSL_ctrl(ssl2, SSL_CTRL_SET_TLSEXT_HOSTNAME, TLSEXT_NAMETYPE_host_name,
| ^
…/boringssl/src/include/openssl/ssl.h:5699:38: note: expanded from macro 'SSL_CTRL_SET_TLSEXT_HOSTNAME'
5699 | #define SSL_CTRL_SET_TLSEXT_HOSTNAME doesnt_exist
| ^
4 errors generated.
```
Change-Id: I5798f17323672d70f75335cea61094457b54466e
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/5769752
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
Most significantly, this includes:
6a848b1a1643 Require a minimum of TLS 1.2 (#1889)
Although Crashpad only uses cpp-httplib in tests, there’s no reason to
taunt fate with this tempting juicy morsel. TLS 1.1 is deprecated
(https://datatracker.ietf.org/doc/html/rfc8996, 2021-03).
This includes a change to util/net/http_transport_test_server.cc to
ensure that the test server, which runs in a child process, continues to
return the full multipart request body as it had in the past. Since
cpp-httplib 7e420aeed361 introduced multipart handling, the raw
multipart wrapper no longer appears in Request::body, but is instead
made available at Request::files. With this change, the test server will
reconstitute the original request body to match the test’s expectations.
Note that this isn’t the only way to serialize the request to be
conveyed back to the test, but it’s the most expedient because it’s what
the test already expects, and because the existing framing already takes
the form of the raw HTTP request.
Change-Id: Ia4adaedff0873976f7cc5be138d78f931165fe4e
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/5753782
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
Since the test is specifically trying to exercise UB by testing the
state of an object it is already destroyed, unpoison the memory to
suppress MSan errors.
Bug: 40222690
Change-Id: I840e944f5e8b39668ac05d8d641fdd5f2e3db5ac
Reviewed-on: https://chromium-review.googlesource.com/c/crashpad/crashpad/+/5716150
Reviewed-by: Mark Mentovai <mark@chromium.org>
