Bug 6665 – Regression(2.055) ICE(cg87.c): static double inside closure

Status
RESOLVED
Resolution
FIXED
Severity
regression
Priority
P2
Component
dmd
Product
D
Version
D2
Platform
x86_64
OS
Linux
Creation time
2011-09-13T15:30:00Z
Last change time
2011-10-22T00:28:07Z
Keywords
ice-on-valid-code
Assigned to
nobody
Creator
iteronvexor

Comments

Comment #0 by iteronvexor — 2011-09-13T15:30:11Z
On a 64-bit Linux machine, with DMD 2.055 This gives: '/Internal error: ../ztc/cg87.c 202' void main(){ auto f = (double m){ static double sum = 0.0; return sum += m * m; }; double[] a = array(map!f(iota(1.0, 25.0, 1.0))); writeln(a); } This one compiles but I get Segmentation fault. void main(){ auto f = (double m){ /* static double sum = 0.0;*/ return m * m; }; double[] a = array(map!f(iota(1.0, 25.0, 1.0))); writeln(a); } This one compiles and runs. void main(){ auto f = (double m){ /* static double sum = 0.0;*/ return m * m; }; double[] a = array(map!f(array(iota(1.0, 25.0, 1.0)))); writeln(a); } Putting everything together, I still get 'Internal error: ../ztc/cg87.c 202' with this one. void main(){ auto f = (double m){ static double sum = 0.0; return sum += m * m; }; double[] a = array(map!f(array(iota(1.0, 25.0, 1.0)))); writeln(a); }
Comment #1 by clugdbug — 2011-09-14T05:16:24Z
Probably a result of the fix for bug 6505. It may be 64 bit specific, I cannot reproduce on Windows. I'm assuming that the imports are: import std.algorithm; import std.array; import std.stdio; import std.range;
Comment #2 by iteronvexor — 2011-09-14T13:03:36Z
Yes, the imports are what you have listed. This might be just a Linux issue. Here are the results for 32-bit Linux: Compiles but gives 'Segmentation fault' void main(){ auto f = (double m){ static double sum = 0.0; return sum += m * m; }; double[] a = array(map!f(iota(1.0, 25.0, 1.0))); writeln(a); } Compiles but gives 'Segmentation fault' void main(){ auto f = (double m){ /* static double sum = 0.0;*/ return m * m; }; double[] a = array(map!f(iota(1.0, 25.0, 1.0))); writeln(a); } Compiles and runs void main(){ auto f = (double m){ /* static double sum = 0.0;*/ return m * m; }; double[] a = array(map!f(array(iota(1.0, 25.0, 1.0)))); writeln(a); } Compiles and runs void main(){ auto f = (double m){ static double sum = 0.0; return sum += m * m; }; double[] a = array(map!f(array(iota(1.0, 25.0, 1.0)))); writeln(a); }
Comment #3 by clugdbug — 2011-09-14T18:15:21Z
Compiling the first case with -inline gives an error: c:\dmd\windows\bin\..\..\src\phobos\std\algorithm.d(423): Error: function D main is a nested function and cannot be accessed from array which is completely wrong -- why does it think main() is a nested function??? So even on Windows this shows a serious problem with the inliner. But the inlining bug isn't a regression, the same error message applies also as far back as 2.035.
Comment #4 by dsimcha — 2011-10-13T10:53:08Z
FWIW, I've found another way to reproduce this bug: import std.range, std.algorithm; double fun(double x) { return x; } void main() { map!(fun)(indexed([1.0, 2], [0, 1])); } Using DMD64 on Linux (This is a 64-bit specific issue, adding -m32 makes it go away. It also only happens with -release.) dmd test.d -release Internal error: ../ztc/cg87.c 202
Comment #5 by dsimcha — 2011-10-13T11:37:07Z
Even more thoroughly reduced test case: void main() { auto foo = Indexed([1.0f, 2, 3], [1, 2, 3]); } @property ref T front(T)(T[] a) { return a[0]; } struct Indexed { float[] _source; int[] _indices; @property float front(float newVal) { return _source[_indices.front] = newVal; } } $ dmd test.d -release Internal error: ../ztc/cg87.c 202
Comment #6 by dsimcha — 2011-10-13T11:38:40Z
BTW, all this stuff works on the latest GDC, so it's definitely not in any code shared between DMD and GDC.
Comment #7 by maximzms — 2011-10-19T07:37:08Z
More investigation. ----- test.d ----- struct Foo { double[2][2] dat; double foo(size_t i, size_t j) { return dat[i][j] = 0; } } void main() { Foo a; } ------------------ Tested on Linux 64bit with dmd 2.055: dmd -m32 test.d -> OK dmd -m64 test.d -> Internal error: ../ztc/cg87.c 202 dmd -m32 -release test.d -> OK dmd -m64 -release test.d -> OK There is no error in any of the following cases: * "dat" is one-dimensional * one of the indexes in Foo.foo is constant (e.g. dat[i][0] = 0) * "int" or "real" used instead of "double" ("float" still produces the error)
Comment #8 by bugzilla — 2011-10-22T00:28:07Z