Commit Briefs
fix gotsysd.conf "listen on" statement failing on non-existent sockets
Reported by Mischa Peters
gotd_accept_paused() is redundant, remove it
The gotd_accept() handler already handles the EV_TIMEOUT case. We do not need a separate function for this case.
prevent gotd looping on unhandled errors raised by accept()
gotd would somehow loop on ENOTSOCK in a case observed by "noodle" on IRC. For now, exit the event loop if an unknown accept error occurs.
gotwebd: rename LOGIN token to separate namespace
There's differences in how yacc(1) works on BSD vs non-BSD systems. As such, the generated code can often clash with tokens which are defined elsewhere. In this case, LOGIN from gotwebd clashes with LOGIN defined in gotwebd.h Simple solution here is to rename the LOGIN token in gotwebd to be GOTWEBD_LOGIN OK @stsp, @op2
make the sockets used for gotwebd IPC use non-blocking I/O as intended
Hopefully fixes a deadlock I observed where the sockets and fcgi processes were stuck in sendmsg(2) trying to send messages to each other.
add an auth process to gotwebd which handles web request authentication
Each gotweb worker now has an auth process sitting in front of it where we process web logins and authenticate requests. The auth process has no access to files other than read-only access to the password database via the "getpw" pledge promise. ok op@
tweak gotwebd deferred accept handling to work more like httpd's
The arbitrary(?) limit of 6 file descriptors per request no longer applies to the sockets process. Change fd reserve limits to match those of httpd such that we can process all connections it throws at us before httpd stops accepting more of them itself. Remove a redundant accept timeout handler function as well. ok op@
fix an issue where histedit would not merge added files correctly
When a file is added during a merge and also already exists on disk as an unversioned file, we mistakenly did not put the file onto the list of changed files which histedit includes in the next commit. This could result in commits which only add new files being dropped as no-op commits. If future commits depend on the file being present this would result in errors during the histedit operation. Otherwise, the file would no longer exist in the edited history. ok op@
use an absolute URL in redirects generated by gotwebd
I am going to be using HTTP::Tiny in upcoming regression tests and this HTTP client fails on redirects sent by gotwebd unless the URL is absolute.
