Re: Support a wildcard in backtrace_functions

On 2024-Mar-06, Bharath Rupireddy wrote:

> +1 for disallowing *foo or foo* or foo*bar etc. combinations.

Cool.

> I think we need to go a bit further and convert backtrace_functions of
> type GUC_LIST_INPUT so that check_backtrace_functions can just use
> SplitIdentifierString to parse the list of identifiers. Then, the
> strspn can just be something like below for each token:
>
> validlen = strspn(*tok,
> "0123456789_"
> "abcdefghijklmnopqrstuvwxyz"
> "ABCDEFGHIJKLMNOPQRSTUVWXYZ");
>
> Does anyone see a problem with it?

IIRC the reason it's coded as it is, is so that we have a single palloc
chunk of memory to free when the value changes; we purposefully stayed
away from SplitIdentifierString and the like. What problem do you see
with the idea I proposed? That was:

> On Thu, Feb 29, 2024 at 4:05 PM Alvaro Herrera <alvherre(at)alvh(dot)no-ip(dot)org> wrote:

> > I think we should tighten this up: an asterisk should be allowed
> > only if it appears alone in the string (short-circuiting
> > check_backtrace_functions before strspn); and let's leave the
> > strspn() call alone.

That means, just add something like this at the top of
check_backtrace_functions and don't do anything to this function
otherwise (untested code):

if (newval[0] == '*' && newval[1] == '\0')
{
someval = guc_malloc(ERROR, 2);
if (someval == NULL)
return false;
someval[0] = '*';
someval[1] = '\0';
*extra = someval;
return true;
}

(Not sure if a second trailing \0 is necessary.)

--
Álvaro Herrera Breisgau, Deutschland — https://www.EnterpriseDB.com/
Voy a acabar con todos los humanos / con los humanos yo acabaré
voy a acabar con todos (bis) / con todos los humanos acabaré ¡acabaré! (Bender)