Closed maxcrees closed 4 years ago
Paging @indutny and also @jasnell, @mcollina and @sam-github.
It seems like a pretty obvious oversight (I caught it immediately while looking at the diff so I'm pretty disappointed no one else did...) so please figure out how to fix this.
Sigh :(.
FWIW, I've marked relevant PR as an ABI change. Looks like the details went missing somewhere in the process. Should we cut 2.9.4
with the revert and then 2.10.0
?
Sorry!
As @indutny mentions, this was clearly an abi breaking change. Unfortunate that that was missed in the release process. A revert plus new 2.10.0 should work I think.
Should we cut 2.9.4 with the revert and then 2.10.0?
Just to be clear, 2.10.0 would contain the ABI change again? With a SONAME bump? What happens when a security vulnerability is found that affects 2.9.4?
I think the change from 7d5c99d can be done in a backwards compatible way. PTAL y'all at #503.
Fixed in 714cbb2 and released as v2.9.4.
Hello,
After spending quite some time debugging quassel-webserver breaking on my latest system update, I traced the issue back to the upgrade of http-parser from 2.9.2 to 2.9.3 without a corresponding rebuild of node itself. Node wasn't rebuilt because the SONAME of http-parser didn't change.
However, it is clear that the changes introduced in 7d5c99d09f6743b055d53fc3f642746d9801479b horribly broke the ABI of libhttp_parser.so. This is due to the changes made to
enum http_errno
andstruct http_parser
(the changes made toenum header_states
are also breaking but I'm assuming this is not ABI). The abi-laboratory report gives a nice summary of this.The problem I encountered was any request to the app would immediately return a
Bad Request
, but I figure this can manifest in any number of ways - for example, the Fedora people saw that it made libgit2 segfault.Please consider cutting a new release with an updated SONAME or otherwise making the changes ABI compatible.
Thanks,
Max