Re: dmd crash help
Don wrote: strtr wrote: Don Wrote: Excellent! I can reproduce it. The D2 version of this code (char[] text = `Hi`.dup;) segfaults in Expression::arraySyntaxCopy() in expression.c, with evidence of memory corruption. The D1 version does *not* segfault there. In fact the code compiles with the debug version of DMD1! I'd love there to be a built debug version in the zip. I think the array literal isn't getting initialized properly. What is up with the specific length of the enum? If you remove one element it compiles just fine. Bugzilla 3769. Very nasty. And it's fixed in svn. Will be in the next release.
Re: dmd crash help
strtr wrote: Don Wrote: Excellent! I can reproduce it. The D2 version of this code (char[] text = `Hi`.dup;) segfaults in Expression::arraySyntaxCopy() in expression.c, with evidence of memory corruption. The D1 version does *not* segfault there. In fact the code compiles with the debug version of DMD1! I'd love there to be a built debug version in the zip. I think the array literal isn't getting initialized properly. What is up with the specific length of the enum? If you remove one element it compiles just fine. Bugzilla 3769. Very nasty.
Re: dmd crash help
Trass3r Wrote: > > Building is straightforward. Just run make -f*youros*.mak. > Should I be offended ? ;) > > > Nope. I just mentioned it cause I experience it very seldom that something > > compiles that painlessly out of the box. Sorry, it was a lame joke. At first glance it looked like self-censoring.
Re: dmd crash help
Should I be offended ? Nope. I just mentioned it cause I experience it very seldom that something compiles that painlessly out of the box.
Re: dmd crash help
Trass3r Wrote: > > Building is straightforward. Just run make -f*youros*.mak. Should I be offended ? > Works flawlessly out of the box for me with dmc on windows. Took me only a few minutes, thanks. But I had to install a compiler to a different language ;)
Re: dmd crash help
I'd love there to be a built debug version in the zip. Building is straightforward. Just run make -f*youros*.mak. Works flawlessly out of the box for me with dmc on windows.
Re: dmd crash help
Don Wrote: > Excellent! I can reproduce it. > The D2 version of this code (char[] text = `Hi`.dup;) segfaults in > Expression::arraySyntaxCopy() in expression.c, with evidence of memory > > corruption. The D1 version does *not* segfault there. In fact the code > compiles with the debug version of DMD1! I'd love there to be a built debug version in the zip. > I think the array literal isn't getting initialized properly. What is up with the specific length of the enum? If you remove one element it compiles just fine. > > > And it's a regression, introduced in DMD1.046. It worked in 1.045. 1.048 wasn't far enough back :)
Re: dmd crash help
Don wrote: strtr wrote: strtr Wrote: dmd(1.048/55/56) crashed on me: AppName: dmd.exe AppVer: 0.0.0.0 ModName: unknown ModVer: 0.0.0.0 Offset: 0002 There is a lot of crash data but I do not know where to start looking. I've also tried finding which part of the code is responsible, but I haven't found it yet. Everything is working again and I have found this piece of code which chokes the compiler. module main; //Digital Mars D Compiler v1.056 //bud_win_3.04.exe -full -cleanup enum ENUM { N0 = 1, //N0 = x; crashes for some values of x, ie. 1 and 10. and not for others ie. 3. N1, N2, N3, N4, N5, N6, N7, N8, N9, N10, N11, N12, N13, N14, N15, N16, N17, N18 } const char[][ ENUM.max + 1 ] ENUM_NAME = [ ENUM.N0 : "N0", ]; void main() { char[] text = `Hi`; switch(text) { //case `N0`://this works case ENUM_NAME[ENUM.N0]://this doesn't //return ENUM.N0; default: break; } } I will post it together with my previous reported enum incomprehensibility bug if it can't be further minimized. Excellent! I can reproduce it. The D2 version of this code (char[] text = `Hi`.dup;) segfaults in Expression::arraySyntaxCopy() in expression.c, with evidence of memory corruption. The D1 version does *not* segfault there. In fact the code compiles with the debug version of DMD1! I think the array literal isn't getting initialized properly. And it's a regression, introduced in DMD1.046. It worked in 1.045.
Re: dmd crash help
strtr wrote: strtr Wrote: dmd(1.048/55/56) crashed on me: AppName: dmd.exe AppVer: 0.0.0.0 ModName: unknown ModVer: 0.0.0.0 Offset: 0002 There is a lot of crash data but I do not know where to start looking. I've also tried finding which part of the code is responsible, but I haven't found it yet. Everything is working again and I have found this piece of code which chokes the compiler. module main; //Digital Mars D Compiler v1.056 //bud_win_3.04.exe -full -cleanup enum ENUM { N0 = 1, //N0 = x; crashes for some values of x, ie. 1 and 10. and not for others ie. 3. N1, N2, N3, N4, N5, N6, N7, N8, N9, N10, N11, N12, N13, N14, N15, N16, N17, N18 } const char[][ ENUM.max + 1 ] ENUM_NAME = [ ENUM.N0 : "N0", ]; void main() { char[] text = `Hi`; switch(text) { //case `N0`://this works case ENUM_NAME[ENUM.N0]://this doesn't //return ENUM.N0; default: break; } } I will post it together with my previous reported enum incomprehensibility bug if it can't be further minimized. Excellent! I can reproduce it. The D2 version of this code (char[] text = `Hi`.dup;) segfaults in Expression::arraySyntaxCopy() in expression.c, with evidence of memory corruption. The D1 version does *not* segfault there. In fact the code compiles with the debug version of DMD1! I think the array literal isn't getting initialized properly.
Re: dmd crash help
strtr Wrote: > dmd(1.048/55/56) crashed on me: > AppName: dmd.exe AppVer: 0.0.0.0 ModName: unknown > ModVer: 0.0.0.0Offset: 0002 > There is a lot of crash data but I do not know where to start looking. > I've also tried finding which part of the code is responsible, but I haven't > found it yet. Everything is working again and I have found this piece of code which chokes the compiler. module main; //Digital Mars D Compiler v1.056 //bud_win_3.04.exe -full -cleanup enum ENUM { N0 = 1, //N0 = x; crashes for some values of x, ie. 1 and 10. and not for others ie. 3. N1, N2, N3, N4, N5, N6, N7, N8, N9, N10, N11, N12, N13, N14, N15, N16, N17, N18 } const char[][ ENUM.max + 1 ] ENUM_NAME = [ ENUM.N0 : "N0", ]; void main() { char[] text = `Hi`; switch(text) { //case `N0`://this works case ENUM_NAME[ENUM.N0]://this doesn't //return ENUM.N0; default: break; } } I will post it together with my previous reported enum incomprehensibility bug if it can't be further minimized.
Re: dmd crash help
strtr wrote: strtr Wrote: dmd(1.048/55/56) crashed on me: I've also tried finding which part of the code is responsible, but I haven't found it yet. I keep forgetting my best method yet : compiling module by module until something strange comes up, or in this case, the compiler chokes. Seems to be related to AAs and invalid code being imported multiple times, or something :) Maybe tomorrow.. Nasty. Please post it to Bugzilla when you track it down. It's really important to get rid of these nightmare bugs.
Re: dmd crash help
strtr wrote: BCS Wrote: Hello Strtr, If you can try it on linux it's not to hard to build a debug version of dmd and run it under the debugger. That will at least give you a filename/line number for the bug report and with a little thinking and a stack trace you can take a guess at what code is causing the failure. I guess the same things can be done on windows (using windbg) but I've never done it there. -- < I didn't even know it was possible to build dmd :) Starting with versions 1.041 and 2.026, the DMD source code is included in the zip file. :) -Lars
Re: dmd crash help
strtr Wrote: > dmd(1.048/55/56) crashed on me: > I've also tried finding which part of the code is responsible, but I haven't > found it yet. I keep forgetting my best method yet : compiling module by module until something strange comes up, or in this case, the compiler chokes. Seems to be related to AAs and invalid code being imported multiple times, or something :) Maybe tomorrow..
Re: dmd crash help
Hello Strtr, -v ended like this : semantic Cg_shader semantic abgr semantic bgra seman This is code within Derelict I think. Strange sudden break.. Stopping mid word makes me think that the output buffer didn't get flushed so something failed after the next to last line, but you don't know how far after. -- <
Re: dmd crash help
BCS Wrote: > Hello Strtr, > > > If you can try it on linux it's not to hard to build a debug version of dmd > and run it under the debugger. That will at least give you a filename/line > number for the bug report and with a little thinking and a stack trace you > can take a guess at what code is causing the failure. I guess the same things > can be done on windows (using windbg) but I've never done it there. > > -- > > < > I didn't even know it was possible to build dmd :)
Re: dmd crash help
Don Wrote: > strtr wrote: > > dmd(1.048/55/56) crashed on me: > > AppName: dmd.exe AppVer: 0.0.0.0 ModName: unknown > > ModVer: 0.0.0.0 Offset: 0002 > > There is a lot of crash data but I do not know where to start looking. > > I've also tried finding which part of the code is responsible, but I > > haven't found it yet. > > Silent crashes are the worst compiler bugs. I'm desperately trying to > eradicate them all, so please submit a test case if you can. > Known bugs which can cause a segfault or stack overflow on Windows DMD1 are: > 3493 > 2315 Won't windows tell me about segfaults and buffer overflows? The same code triggered the Data Execution Prevention on a different machine (also winXP) > > The crash data doesn't tell you anything useful. > Try compiling with dmd -v, and it should give you a little bit of a hint > of what it was doing. > -v ended like this : semantic Cg_shader semantic abgr semantic bgra seman This is code within Derelict I think. Strange sudden break..
Re: dmd crash help
Hello Strtr, dmd(1.048/55/56) crashed on me: AppName: dmd.exe AppVer: 0.0.0.0 ModName: unknown ModVer: 0.0.0.0 Offset: 0002 There is a lot of crash data but I do not know where to start looking. I've also tried finding which part of the code is responsible, but I haven't found it yet. If you can try it on linux it's not to hard to build a debug version of dmd and run it under the debugger. That will at least give you a filename/line number for the bug report and with a little thinking and a stack trace you can take a guess at what code is causing the failure. I guess the same things can be done on windows (using windbg) but I've never done it there. -- <
Re: dmd crash help
strtr wrote: dmd(1.048/55/56) crashed on me: AppName: dmd.exe AppVer: 0.0.0.0 ModName: unknown ModVer: 0.0.0.0 Offset: 0002 There is a lot of crash data but I do not know where to start looking. I've also tried finding which part of the code is responsible, but I haven't found it yet. Silent crashes are the worst compiler bugs. I'm desperately trying to eradicate them all, so please submit a test case if you can. Known bugs which can cause a segfault or stack overflow on Windows DMD1 are: 3493 2315 The crash data doesn't tell you anything useful. Try compiling with dmd -v, and it should give you a little bit of a hint of what it was doing.