Previously, when the --unlink flag was passed to CrashReporter, it
unlinked the coredump file immediately after reading it.
This change makes it so the coredump file is deleted when CrashReporter
exits.
This was a nice idea in theory, but in practice it makes big crashes
(e.g WebContent) even more CPU intensive. Let's disable this for now
(but keep the ability for CrashReporter to open compressed coredumps.)
Before this patch, this is what would happen after something crashed:
1. CrashDaemon finds a new coredump in /tmp
2. CrashDaemon compresses the new coredump (gzip)
3. CrashDaemon parses the uncompressed coredump and prints a backtrace
4. CrashDaemon launches CrashReporter
5. CrashReporter parses the uncompressed coredump (again)
6. CrashReporter unlinks the uncompressed coredump
7. CrashReporter displays a GUI
This was taking quite a long time when dealing with large programs
crashing (like Browser's WebContent processes.)
The new flow:
1. CrashDaemon finds a new coredump in /tmp
2. CrashDaemon mmap()'s the (uncompressed) coredump
3. CrashDaemon launches CrashReporter
4. CrashDaemon goes to sleep for 3 seconds (hack alert!)
5. CrashReporter parses the (uncompressed) coredump
6. CrashReporter unlinks the (uncompressed) coredump
7. CrashReporter displays a GUI
8. CrashDaemon wakes up (after step 4)
9. CrashDaemon compresses the coredump (gzip)
TL;DR: we no longer parse the coredumps twice, and we also prioritize
launching the CrashReporter GUI immediately when a new coredump shows
up, instead of compressing and parsing it in CrashDaemon first.
The net effect of this is that you get a backtrace on screen much
sooner. That's pretty nice. :^)
For compressed coredumps, CrashReporter's argv were in this order:
CrashReporter path --unlink
Core::ArgsParser doesn't like that at all and would immediately exit
from main(), causing the crash reporter to never display.
With the new InodeWatcher API, the old style of creating a watcher per
inode will no longer work. Therefore the FileWatcher API has been
updated to support multiple watches, and its users have also been
refactored to the new style. At the moment, all operations done on a
(Blocking)FileWatcher return Result objects, however, this may be
changed in the future if it becomes too obnoxious. :^)
Co-authored-by: Gunnar Beutner <gunnar@beutner.name>
SPDX License Identifiers are a more compact / standardized
way of representing file license information.
See: https://spdx.dev/resources/use/#identifiers
This was done with the `ambr` search and replace tool.
ambr --no-parent-ignore --key-from-file --rep-from-file key.txt rep.txt *
Most coredumps contain large amounts of consecutive null bytes and as
such are a prime candidate for compression.
This commit makes CrashDaemon compress files once the kernel finishes
emitting them, as well as adds the functionality needed in LibCoreDump
to then parse them.
(...and ASSERT_NOT_REACHED => VERIFY_NOT_REACHED)
Since all of these checks are done in release builds as well,
let's rename them to VERIFY to prevent confusion, as everyone is
used to assertions being compiled out in release.
We can introduce a new ASSERT macro that is specifically for debug
checks, but I'm doing this wholesale conversion first since we've
accumulated thousands of these already, and it's not immediately
obvious which ones are suitable for ASSERT.
This wrapper abstracts the watch_file setup and file handling, and
allows using the watch_file events as part of the event loop via the
Core::Notifier class.
Also renames the existing DirectoryWatcher class to BlockingFileWatcher,
and adds support for the Modified mode in this class.
