fix for rt-43537: undefined values passed to the multi functions no long...

[mcmillhj]

This patch fixes rt-43537:

Calls to set_multi that contained an undefined value caused segmentation faults. The root cause of this was the de-referencing of a call to av_fetch() which returned a NULL pointer.

To fix this a new function, safe_av_fetch(), was added to Fast.xs that checks the result of av_fetch for a NULL pointer or a pointer that does not pass an SvOK() check. If such a value is encountered safe_av_fetch() croaks.

A new test t/malformed.t was added to test this new functionality. It adds three values to the server, then submits a series calls to set_mutli() with invalid arguments (in varying positions). It verifies that none of these calls changes the values of the keys on the server.

[mcmillhj]

My mistake, I even had that on a list of things to do and I still forgot. Is aTHX_ a requirement? I am surprised it built properly if it is.

[kroki]

aTHX_ is an argument (hence "a", underscore suggests a comma) for a pTHX_ parameter (hence "p"). You need aTHX_ because you have pTHX_, and you need pTHX_ because when Perl is built with the support for multi-threading av_fetch() (and friends) is a macro that expands to a function call with an extra argument of thread context. I.e. pTHX_ expands to "ThreadContextType *ctx," (actual names are different), aTHX_ expands to "ctx,", and av_fetch(a,b,c) expands to av_fetch(ctx,a,b,c), thus passing around current thread context "ctx". Before the commit 13c9216 we haven't defined PERL_NO_GET_CONTEXT and so didn't have pTHX_/aTHX_, but then av_fetch(a,b,c) had expanded to av_fetch(get_current_ctx(),a,b,c) and there were too many get_current_ctx() calls, which bothered some.

Presumably on OSX you have single-threaded Perl, hence pTHX_/aTHX_ expand to empty strings, and av_fetch(a,b,c) expands to av_fetch(global_ctx_var,a,b,c), and so you don't have compilation problem. But on Debian you had an error yourself ;).