[WPE][GTK] Null pointer dereference when child process exits immediately

• Component: WebKitGTK
• Status: RESOLVED FIXED

[Michael Catanzaro]

In https://github.com/flathub/org.gnome.Epiphany/issues/21 we discovered that when the child process exits immediately after it is spawned, g_subprocess_get_identifier() will return nullptr. In this case, we should crash cleanly with SIGABRT via g_error(), rather than crashing with a null pointer dereference inside g_ascii_strtoll(). SIGABRT is much nicer than SIGSEGV and indicates that we really do want to crash here, whereas SIGSEGV is just a bug.

[Michael Catanzaro]

Created attachment 425210 [details]
Patch

[Michael Catanzaro]

The crashing test is TestWebsiteData -p /webkit/WebKitWebsiteData/configuration:

ERROR:../../Tools/TestWebKitAPI/Tests/WebKitGLib/TestWebsiteData.cpp:220:void testWebsiteDataConfiguration(WebsiteDataTest*, gconstpointer): 'test->fetch(persistentCaches)' should be nullptr
Bail out! ERROR:../../Tools/TestWebKitAPI/Tests/WebKitGLib/TestWebsiteData.cpp:220:void testWebsiteDataConfiguration(WebsiteDataTest*, gconstpointer): 'test->fetch(persistentCaches)' should be nullptr

My patch causes this test to fail on EWS, but the test crashes for me locally in the same way without my change. I can only imagine what's going wrong with the website data....

[Carlos Garcia Campos]

I think the api test failure is unrelated.

[Michael Catanzaro]

Comment on attachment 425210 [details]
Patch

Let's find out, then....

[EWS]

Committed r275605: <https://commits.webkit.org/r275605>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 425210 [details].