http://d.puremagic.com/issues/show_bug.cgi?id=884

           Summary: Segfault in recursive template
           Product: D
           Version: 1.001
          Platform: PC
        OS/Version: Linux
            Status: NEW
          Severity: normal
          Priority: P2
         Component: DMD
        AssignedTo: bugzilla@digitalmars.com
        ReportedBy: fvbommel@wxs.nl


The following code segfaults DMD v1.001 (Linux, don't know about Windows)
-----
template Inner(Ts...) {

    const Inner = .Inner!((Ts));
}



auto x = Inner!();
-----

Just so you people don't think I'm crazy for trying to compile that ;), let me just state for the record that above code is *way* cut down from much more complicated code. I removed everything I could while retaining the segfault.

Above code is probably invalid. The original code may or may not have been valid, I'm not sure.

The original code also segfaulted on v1.00, and perhaps earlier (not sure when I started working on it).

Some notes I made while cutting it down:
* The Ts in the initializer was originally sliced with [1..$], with the same
result (if filled with some value parameters). Only in 'static if (Ts.length >
0)', of course.
* The initializer expression in general used to be a bit more complicated :p.
* Braces around Ts were initially added to avoid "tuple Ts is used as a type",
later avoided "tuple is not a valid template value argument" and currently need
to be retained to eliminate "circular reference to 'Inner'", which is pretty
obvious in the current code :). Since the code that produced those errors
didn't segfault DMD, they stayed in.
* The Inner member originally had type T[] (T being the template parameter of
an enclosing template), but the compiler segfaulted with void[] as well.


-- 

http://d.puremagic.com/issues/show_bug.cgi?id=884


thomas-dloop@kuehne.cn changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         OS/Version|Linux                       |All




------- Comment #1 from thomas-dloop@kuehne.cn  2007-04-05 05:27 -------
Added to DStress as http://dstress.kuehne.cn/nocompile/t/templatde_58_A.d http://dstress.kuehne.cn/nocompile/t/templatde_58_B.d http://dstress.kuehne.cn/nocompile/t/templatde_58_C.d http://dstress.kuehne.cn/nocompile/t/templatde_58_D.d


-- 

http://d.puremagic.com/issues/show_bug.cgi?id=884


matti.niemenmaa+dbugzilla@iki.fi changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |matti.niemenmaa+dbugzilla@ik
                   |                            |i.fi
           Keywords|                            |ice-on-valid-code
           Platform|PC                          |All




------- Comment #2 from matti.niemenmaa+dbugzilla@iki.fi  2008-08-17 04:25 -------
Another testcase, somewhat different (just ran into this in 1.034 on Windows):

template Ins() { const Ins = Ins!(Ins); }
alias Ins!() x;

This was reduced from the following, which should be valid but loops infinitely (added ice-on-valid-code, since the code is valid, just divergent):

template Ins(alias x) { const Ins = Ins!(Ins); }
alias Ins!(Ins) x;

Doing the following instead makes DMD correctly report a circular reference error:

template Ins() { const Ins = Ins!(); }
alias Ins!() x;


-- 

http://d.puremagic.com/issues/show_bug.cgi?id=884





------- Comment #3 from clugdbug@yahoo.com.au  2009-04-02 15:30 -------
The first of Matti's cases is OK on DMD2.027.
The second is segfaulting in optimize.c,
Expression *expandVar(int result, VarDeclaration *v),
in the line:

        Type *tb = v->type->toBasetype();

type may be null. This can trivially be changed into an ICE, rather than a segfault, not that's much of an improvement.


-- 

http://d.puremagic.com/issues/show_bug.cgi?id=884





------- Comment #4 from maxmo@pochta.ru  2009-04-04 08:33 -------
Seems like dmd uses too much field access for order-independent language.


-- 

http://d.puremagic.com/issues/show_bug.cgi?id=884





------- Comment #5 from clugdbug@yahoo.com.au  2009-04-24 02:48 -------
Created an attachment (id=336)
 --> (http://d.puremagic.com/issues/attachment.cgi?id=336&action=view)
Patch against DMD2.029

This patch stops Matti's case from segfaulting, and gives an error message on the offending line. The DStress cases are no longer segfaulting, they ICE instead.


-- 

http://d.puremagic.com/issues/show_bug.cgi?id=884


clugdbug@yahoo.com.au changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |patch




------- Comment #6 from clugdbug@yahoo.com.au  2009-04-24 02:51 -------
Note that this patch doesn't close the original bug, it just fixes the
(different) bug from Matti. With this patch applied, the bug title can become
"ICE(template.c) in recursive template".


-- 

http://d.puremagic.com/issues/show_bug.cgi?id=884


Don <clugdbug@yahoo.com.au> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|patch                       |
             Status|NEW                         |RESOLVED
                 CC|                            |clugdbug@yahoo.com.au
         Resolution|                            |FIXED
            Summary|Segfault in recursive       |ICE(template.c) in
                   |template                    |recursive template




--- Comment #7 from Don <clugdbug@yahoo.com.au>  2009-05-14 05:15:54 PDT ---
All segfaults are fixed in D2.030 and 1.045. The original test case still ICEs, though, but I'm moving that to issue 2884 (which seems to be the same bug).

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------