Bug 7926 – stack overflow on recursive string mixin

$ cat > x.d enum string a = "mixin(a);"; mixin(a); $ dmd x.d Segmentation fault Tested on recent DMD >2.059 from master branch of GIT repository: HEAD is at 778aea7fa8fe9ce3be6f70c7c99b69fd17d6e0b1

On 2.059 Win32 PS E:\DigitalMars\dmd2\samples> dmd -c bug.d Stack overflow which I guess is what's to be expected.

I don't think this is a bug. The infinite recursive expansion is the intended behavior of the code. We can write arbitrary code in this kind. // mutual recursion version enum string a = "mixin(b);"; enum string b = "mixin(a);"; mixin(a); // infinite code bloating version string a(string num = "1") { string inc(string num) { return (num.length == 0 ? "1" : num[$-1] == '9' ? inc(num[0..$-1]) ~ "0" : num[0..$-1] ~ cast(char)(num[$-1] + 1)); } return `enum a`~num~` = a("`~inc(num)~`"); `~ `/*pragma(msg, a`~num~`);*/` ~ `mixin(a`~num~`);`; } mixin(a);

Still, the diagnostics can be better than just "Error: out of memory". At least showing a [reasonably-sized tail of] what consumed all that memory. On the other hand, it is tricky to log when you already have no memory, and adding logging information for such cases will probably increase the memory footprint in the general case, too.

(In reply to Ivan Kazmenko from comment #3) > Still, the diagnostics can be better than just "Error: out of memory". At > least showing a [reasonably-sized tail of] what consumed all that memory. That's right. "Segmentation fault" gives no clue on what's went bad, and it happens at compile time, so it's pretty hard to find the source of the problem. You might know your own sources well, but digging into the compiler sources is not an option for most, even experienced, developers.