Comment #0 by Jesse.K.Phillips+D — 2011-03-30T21:56:47Z
Created attachment 936
Example code
I've been working toward a CSV parser that works efficiently with an input
range. I'm testing it with string and have come to the conclusion there is a
bug in DMD. But I'm having some issue with reducing this code further[1].
The issue manifests itself when I pass a string to a function, which is stored
in a range that produces a range that stores a pointer to the string, which
modifies the dereferenced string pointer.
The pointer is no longer valid by the time popFront is called. A segmentation
fault is produced on line 95 inside countUntil (the string is garbage).
1. https://gist.github.com/895807
Comment #1 by braddr — 2011-03-30T22:27:12Z
I haven't looked at this under a debugger, but what I suspect is happening is this:
In csvText, you're creating a temporary RecordList which is copied on return if NRVO isn't happening. The pointer inside recordRange points to the temporary. After the copy, the temporary goes away and the pointer is invalid.
Test this by printing the address of RecordList._input inside the RecordList ctor and again in RecordRange.front. If I'm right, the two addresses will differ and the pointer value inside recordRange._input will be the first value.
Assuming all of the above is right, the fix is to add a postblit to fix up the pointer.
Comment #2 by Jesse.K.Phillips+D — 2011-03-31T07:11:27Z
Oh, that makes sense. Adding a postblitz is providing a resolution to this issue. My CSV parser is still segv, but it is in another location. I'll close this as it obviously isn't a bug.