Hello, Good stuff here, I also have a bunch of bugfix commits that ended up not having a test because of the need for a debugger or other interaction, so let's move forward.
I think the docs (and the macro/function naming) describe things backwards. In my mind, it is INJECTION_POINT_RUN() that creates the injection point; then InjectionPointCreate() attaches something to it. So I would rename the macro to just INJECTION_POINT() and the function to InjectionPointAttach(). This way you're saying "attach function FN from library L to the injection point P"; where P is an entity that is being created by the INJECTION_POINT() call in the code. You named the hash table InjectionPointHashByName, which seems weird. Is there any *other* way to locate an injection point that is not by name? In this patch, injection points are instance-wide (because the hash table is in shmem). As soon as you install a callback to one point, that callback will be fired in every session. Maybe for some tests this is OK (and in particular your TAP tests have them attached in one ->safe_psql call and then they hit a completely different session, which wouldn't work if the attachments were process-local), but maybe one would want them limited to some specific process. Maybe give an optional PID so that if any other process hits that injection point, nothing happens? -- Álvaro Herrera Breisgau, Deutschland — https://www.EnterpriseDB.com/ "La rebeldía es la virtud original del hombre" (Arthur Schopenhauer)
