On Wed, Dec 17, 2008 at 4:33 PM, Michael Nordman <[email protected]> wrote: > On Wed, Dec 17, 2008 at 4:11 PM, Michael Nordman <[email protected]> wrote: >> On Wed, Dec 17, 2008 at 4:01 PM, Nigel Tao <[email protected]> wrote: >>> This probably is just a bug. It sounds reasonable that a marshal should >>> either >>> completely succeed or completely fail. >>> >>> Out of curiousity, is there a specific real-world scenario that's motivating >>> this? >> >> This came out of fixing crashing bug http://b/issue?id=1523881 which >> amounted to testing if NewObject and NewArray return null. This code >> was right next to it so i noticed it. >> >> I'm not certain of the exact circumstances that cause NewObject and >> NewArray to fail. If it is sporadic, message recipients could receive >> partial messages (malformed arrays/objects), which could cause bogus >> data to be persisted in a database depending of what the recipient >> does with the message. >> > > Worth mentioning that in all occurances of the crash that i've seen > stack traces for... the NewObject/NewArray method returned null when > creating the top-level object rather than a nested object. So nothing > would come out instead of a malformed something in those cases.
Also worth mentioning that i have yet to filter chrome crash reports to select only those that crashed in gears.dll. Its difficult to find these needles in that haystack. > >>> >>> Can you add a test case to test/testcases/workerpool_message_body_tests.js? >> >> I'm not sure how to induce this failure mode? >> >>> >>> -- >>> To respond, reply to this email or visit >>> http://mondrian.corp.google.com/9455628 >>> >> >
