Hi On Mon, Aug 19, 2024 at 10:12 PM Tom Lane <t...@sss.pgh.pa.us> wrote: > > We have had multiple instances of code "return"ing out of a PG_TRY, > so I fully agree that some better way to detect that would be good. > But maybe we ought to think about static analysis for that.
I have some static analysis scripts for detecting this kind of problem (of mis-using PG_TRY). Not sure if my scripts are helpful here but I would like to share them. - A clang plugin for detecting unsafe control flow statements in PG_TRY. https://github.com/higuoxing/clang-plugins/blob/main/lib/ReturnInPgTryBlockChecker.cpp - Same as above, but in CodeQL[^1] script. https://github.com/higuoxing/postgres.ql/blob/main/return-in-PG_TRY.ql - A CodeQL script for detecting the missing of volatile qualifiers (objects have been changed between the setjmp invocation and longjmp call should be qualified with volatile). https://github.com/higuoxing/postgres.ql/blob/main/volatile-in-PG_TRY.ql Andres also has some compiler hacking to detect return statements in PG_TRY[^2]. [^1]: https://codeql.github.com/ [^2]: https://www.postgresql.org/message-id/20230113054900.b7onkvwtkrykeu3z%40awork3.anarazel.de
