Fix victim lookup after op mode binds

[tkimball83]

Found by: tkimball83
Patch by: tkimball83
Fixes:

One-line summary:

Initialize the mode target's nick/userhost before refreshing its user flags after op and halfop Tcl mode binds.

Additional description (if needed):

got_op() and got_halfop() pass the local s buffer to modebind_refresh() without initializing it. modebind_refresh() then uses that buffer as the host argument to lookup_user_record() while refreshing the target's flag record.

As a result, the refreshed victim flags can belong to no user or the wrong user. With +bitch, a recognized bot that has global op can then be treated as unauthorized and deopped immediately after receiving op.

Populate s from the channel member before invoking the Tcl mode bind, matching the existing pattern in got_deop() and the voice-mode handling.

Test cases demonstrating functionality (if applicable):

• Reproduced with two instances of the official eggdrop:1.10.1 Linux image in a ten-bot network using +bitch.
• Before testing, both rebuilt bots' current IRC masks were confirmed saved exactly once in every bot's userfile, and the channel was allowed to finish NAMES/WHO synchronization.
• After each controlled deop, the target was correctly reopped, but a stock bot issued an unsolicited deop of the recognized target about one second later. This repeated throughout the test.
• Repeated the same post-synchronization mode churn with a 1.10.1 image containing only these two added lines: 20 controlled deop/reop cycles completed without an unsolicited deop.
• Current develop compiles successfully on macOS arm64 with ./configure --disable-tls && make config && make -j4.

[michaelortmann]

Thank you alot for this PR.

The bug is a regression introduced in eggdrop 1.10.0 via #1585.

Can you please take a look at got_dehalfop() also? Looks like that function also needs fixing.

[tkimball83]

Updated the PR to fix got_dehalfop() as suggested. It now initializes the victim host string before calling modebind_refresh(), and I removed the unused s1 buffer/assignment from that function as dead code.

The updated change passes the debug build and the full integration suite: 159 passed, 0 failed.

[michaelortmann]

LGTM

[thommey]

nice find and fix, thank you!