www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.bugs - [Issue 5595] New: Compiler crash on heavy std.algorithm use

reply d-bugmail puremagic.com writes:
http://d.puremagic.com/issues/show_bug.cgi?id=5595

           Summary: Compiler crash on heavy std.algorithm use
           Product: D
           Version: D2
          Platform: Other
        OS/Version: Windows
            Status: NEW
          Keywords: ice-on-valid-code
          Severity: regression
          Priority: P2
         Component: DMD
        AssignedTo: nobody puremagic.com
        ReportedBy: dsimcha yahoo.com


--- Comment #0 from David Simcha <dsimcha yahoo.com> 2011-02-16 07:39:21 PST ---
The following code crashes the compiler on Windows using the latest beta of
2.052.  It works on 2.051.

import std.algorithm, std.math, std.stdio, std.conv;

// std.datetime isn't used but needs to be imported to reproduce the bug.
import std.datetime;

void invert(double[][] mat) {
    foreach(i, row; mat) {
        double absMax = map!(abs)(row).front;
    }
}

void readExp() {
    auto handle = File("foo.txt");
    auto lines = handle.byLine();
    auto ls = lines.front().splitter('\t');
    if(!ls.empty) {
        auto floats = map!(to!float)(ls).front;
    }
}

void main(string[] args) {}

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
Feb 16 2011
next sibling parent d-bugmail puremagic.com writes:
http://d.puremagic.com/issues/show_bug.cgi?id=5595


Don <clugdbug yahoo.com.au> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |clugdbug yahoo.com.au


--- Comment #1 from Don <clugdbug yahoo.com.au> 2011-02-16 07:58:26 PST ---
Confirmed. I'm on it.

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
Feb 16 2011
prev sibling next sibling parent d-bugmail puremagic.com writes:
http://d.puremagic.com/issues/show_bug.cgi?id=5595



--- Comment #2 from Don <clugdbug yahoo.com.au> 2011-02-16 08:28:43 PST ---
The regression was caused by a Phobos change, which has triggered a compiler
segfault in interpret.c.
In attempting to instantiate map!(to), 'to' is a symbol with no identifier.
(.ident is NULL).
The segfault can be prevented in dsymbol.c, line 70, by checking for a null
identifier. But without knowing why we're getting a null identifier, I can't
recommend this as a valid patch.

int Dsymbol::equals(Object *o)
{   Dsymbol *s;

    if (this == o)
        return TRUE;
    s = (Dsymbol *)(o);
-    if (s && ident->equals(s->ident))
+    if (s && s->ident && ident->equals(s->ident))
        return TRUE;
    return FALSE;
}

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
Feb 16 2011
prev sibling next sibling parent d-bugmail puremagic.com writes:
http://d.puremagic.com/issues/show_bug.cgi?id=5595



--- Comment #3 from Don <clugdbug yahoo.com.au> 2011-02-16 11:51:32 PST ---
Slightly reduced test case:
-----
import std.algorithm;
import std.datetime;
import std.math;

void foo(A)(A x) {}

void readExp() {
    int xxx;
    double[] yyy;
    map!(abs)(yyy);
    map!(foo)(xxx);
}

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
Feb 16 2011
prev sibling next sibling parent d-bugmail puremagic.com writes:
http://d.puremagic.com/issues/show_bug.cgi?id=5595



--- Comment #4 from Don <clugdbug yahoo.com.au> 2011-02-16 13:12:13 PST ---
Reduced test case shows it involves overload sets of templates. Segfaults as
far back as 2.012. Definitely not a regression.

==== test1.d ===
void bar(D)(D x) {}

==== test2.d ===
void bar(N)(N x) {}

==== test0.d ===
import test1, test2;

template map(fun...)
{
    void map(R)(R r) {}
}

void foo(A)(A x) {}

void baz() {
    int xxx;
    char yyy;
    map!(bar)(yyy);
    map!(foo)(xxx);
}

-------
dmd test0
<segfault>

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
Feb 16 2011
prev sibling next sibling parent d-bugmail puremagic.com writes:
http://d.puremagic.com/issues/show_bug.cgi?id=5595



--- Comment #5 from Don <clugdbug yahoo.com.au> 2011-02-16 14:23:24 PST ---
Another reduction, uses test1.d and test2.d from the previous comment. It's
clear that IFTI is not required. The issue is instantiating a tuple with an
overload set. If you change the baz(F...) to baz(F) you get:
test0.d(7): Error: template instance baz!(__anonymous) does not match template
declaration baz(F)
which shows that the overload set really doesn't have an identifier.
Also, you can replace the (F...) with (alias F), and still get the segfault. So
it is not a problem with tuples; it's a problem with overload sets of
templates.

=====

import test1, test2;

void foo(A)() {}

template baz(F...)
{
    enum int baz = 1;
}

static assert(baz!bar == baz!foo);

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
Feb 16 2011
prev sibling next sibling parent d-bugmail puremagic.com writes:
http://d.puremagic.com/issues/show_bug.cgi?id=5595


Don <clugdbug yahoo.com.au> changed:

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


--- Comment #6 from Don <clugdbug yahoo.com.au> 2011-02-16 23:40:07 PST ---
I think my initial patch was basically correct. Overload sets are not equal to
other Objects.

----
int Dsymbol::equals(Object *o)
{   Dsymbol *s;

    if (this == o)
        return TRUE;
    s = (Dsymbol *)(o);
-    if (s && ident->equals(s->ident))
+    // Overload sets don't have an ident
+    if (s && ident && s->ident && ident->equals(s->ident))
        return TRUE;
    return FALSE;
}

The interesting thing, which I hadn't realized before, is that a template alias
parameter can be an overload set.
The question this raises is, are there other places in the code where it's
assumed that every DSymbol has an ident?

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
Feb 16 2011
prev sibling parent d-bugmail puremagic.com writes:
http://d.puremagic.com/issues/show_bug.cgi?id=5595


Walter Bright <bugzilla digitalmars.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
                 CC|                            |bugzilla digitalmars.com
         Resolution|                            |FIXED


--- Comment #7 from Walter Bright <bugzilla digitalmars.com> 2011-02-17
16:19:28 PST ---
https://github.com/D-Programming-Language/dmd/commit/f8ed1b3a6d8f1391962111a6d06783cf09bfd90a

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
Feb 17 2011