alley: static destructors

[dtzWill]

It appears these are not currently executed in all cases when running a program via the JIT strategy using alley. Example of this can be seen in the dtor_exit.ll test, changing this line:

allvm-tools/test/lit/dtor_exit.ll

Line 9 in 22a3c91

; RUNX: alley %t/test |& FileCheck %s

to have RUN instead of RUNX. I have not investigated this further to determine if there are instances these do execute, but this should be done and issue updated (or new issues filed) appropriately. I've not thought about it hardly at all but offhand don't see an immediately accessible way to handle this. Thoughts/ideas/welcome.

[dtzWill]

Seeing as I use alley for a significant amount of my daily software.... this is somewhat surprising/interesting. "Global destructors are overrated, the ALLVM story" (kidding!). Almost certainly due to my personal needs tending towards simpler CLI or client utilities, as opposed to server or other software where failing to release resources (or failure to do so gracefully) has a more significant impact. Anyway as long as no headlines such as above are entertained it's still "neat" to find this less necessary than other details.

Note none of our published numbers involve sizes of non-AOT'd allexe's (so no alley JIT) and of course similarly don't included measurements of execution time omitting such destructors :). In case anyone was concerned 👍.