[Issue 1933] New: Delimited string constants can cause segfault

d-bugmail at puremagic.com d-bugmail at puremagic.com
Thu Mar 20 00:12:58 PDT 2008


http://d.puremagic.com/issues/show_bug.cgi?id=1933

           Summary: Delimited string constants can cause segfault
           Product: D
           Version: 2.007
          Platform: PC
        OS/Version: Windows
            Status: NEW
          Severity: normal
          Priority: P2
         Component: DMD
        AssignedTo: bugzilla at digitalmars.com
        ReportedBy: unknown at simplemachines.org


The documentation (which is somewhat unclear) seems to indicate that the
following is incorrect:

q"
EOS
...
EOS"

If you try this, it will cause a segfault.  The reason is that once we leave
the switch in delimitedStringConstant(), c is '\n' (because of the break.) 
This ends up meaning delimleft and delimright are '\n', as best I can reason
(haven't spent time debugging code yet.)

Simple solution to me seems like either adding an if inside |if (delimleft ==
0)| to check that c is not whitespace, or adding a branch before it to ignore
whitespace (depending on which is the correct interpretation of the spec.)

Also would be nice if the spec were clearer (e.g. "the delimiter, whether a
character or identifier, must immediately follow the " without any whitespace. 
If the delimiter is an identifier, it must similarly immediately proceed the
terminating ".)

I realize my D version is a bit outdated, but this diff shows the problem can't
have been fixed yet for 2.012:

http://www.dsource.org/projects/dsss/changeset?new=branches%2Fdmdfe-2.0%2Flexer.c%40883&old=branches%2Fdmdfe-2.0%2Flexer.c%40838

Not sure if you care for my patches, but if I have time I'll try to look into
this more and attach a patch (since I'm fairly confident the whitespace isn't
supposed to be allowed.)

Thanks.  I noticed this because I was writing my own lexer, not because I was
actually using the heredocs... I have to say token and delimited strings
complicated my lexer more than I expected...


-- 



More information about the Digitalmars-d-bugs mailing list