iostream et STL

→ **MyCalcs** profile · de **Excale** » 30 Mar 2014, 09:55

Maybe the malloc() for storing a 7 MB large giac.tns fails

Il est possible que le malloc() pour caser giac.tns rate sur 7Mo.

Can you see "ndless_load: can't malloc" or "Failed to alloc binary memory" in the terminal output

Est ce que tu vois: "ndless_load: can't malloc" ou bien "Failed to alloc binary memory" en console?

If you do, then we'll have to be smart, the CX has much more than enough RAM (but it needs to be contiguous)

Si oui, il va falloir ruser un peu, la Nspire CX a largement assez de RAM libre (le tout étant de la rendre contiguë).

EDIT Lionel: English spoken in this topic

EDIT²: I can run it with Ndless 3.1 and type things. It crashes when I press enter (which is probably what it is supposed to do since Ndless 3.1 may be missing some of features you use)

→ **MyCalcs** profile · de **parisse** » 30 Mar 2014, 10:41

Yes, if I run it without the debugger, I get ndless_load: can't malloc. Which means that the reboot is probably a bad interaction with the malloc and gdb.
So I should definitively try it with the 3.1 OS, since it reaches the console input at the beginning of main(), then it crashes while calling giac parser, there is hope I can debug it and find what fails.

→ **MyCalcs** profile · de **Excale** » 30 Mar 2014, 11:13

parisse a écrit:Yes, if I run it without the debugger, I get ndless_load: can't malloc. Which means that the reboot is probably a bad interaction with the malloc and gdb.

The file is probably just too big and malloc cannot find enough contiguous memory for it (OS 3.1 is smaller so it leaves more free RAM for programs). If it happens to be that issue for sure, we can patch Ndless and ask the OS to malloc enough room for you soon enough in the OS boot sequence so it doesn't fail (but that's not a very good solution, there must be better. This could at least allow you to test on OS 3.6.).

→ **MyCalcs** profile · de **ExtendeD** » 30 Mar 2014, 11:53

Excale, do you mean pre-reserving memory space for Ndless programs? I'm not sure it's a good idea as this space would be lost for OS use. And how much should be reserved this way?
Anyway a 7M allocation on 64M system is much. Would bFLT library support help? or nl_exec() to run another program? ( http://hackspire.unsads.com/wiki/index. ... _functions )

→ **MyCalcs** profile · de **parisse** » 30 Mar 2014, 12:28

I have set the emulator with the 3.1 os and ndless, and I can start a debugging session, but as soon as I try to debug in the parser code, the system reboots. More precisely when I enter s in the gdb console for stepping in giac_yyparse, then it reboots. I suspect that the memory is almost full, and a memory allocation fails at the begin of the parser code.

→ **MyCalcs** profile · de **Vogtinator** » 30 Mar 2014, 12:31

ExtendeD a écrit:Excale, do you mean pre-reserving memory space for Ndless programs? I'm not sure it's a good idea as this space would be lost for OS use. And how much should be reserved this way?
Anyway a 7M allocation on 64M system is much. Would bFLT library support help? or nl_exec() to run another program? ( http://hackspire.unsads.com/wiki/index. ... _functions )

I don't think that's much actually. It's just a bit more than a tenth of 64MiB. Did you compile giac with "-Os"? And you need "-fpic -mlong-calls -fno-exceptions" (for every file, also ustl and libtommath) if you're not using nspire-g++, otherwise some function calls go into the wrong direction...

parisse a écrit:Is there some howto that describes what needs to be done? Or even better a tarball of a working toolchain?
On my side, I could compile giac with ustl (modified), nspire-io and os.h, the resulting binary is 7M large, but it does not work. I tried to set a breakpoint at the beginning of main, but reboot happens before, it happens during static initializations, despite the fact I turned on all flags to compile giac without static object initializations. And I can't find how to set a breakpoint before main.

Clone my fork on github and run "build_toolchain.sh" in the ndless-sdk/toolchain directory. Uploading my toolchain takes too long, over 70MiB compressed. You should be able to compile it without os.h. Does it work then?
The easiest way to set a breakpoint at program startup is by adding "bkpt" in crt0.S in ndless-sdk/system. And set the "/W" flag when running nspire_emu, sometimes it helps debugging.

I haven't got much time today, but I'll try to compile giac as well.

→ **MyCalcs** profile · de **parisse** » 30 Mar 2014, 13:00

I make another try with some optimization template code disabled in giac. I'll try newlib later (but I'm afraid the code will take more memory).
I can't see the option -mlong-call in nspire-g++ shell script? Two other questions: can I use -mthumb in nspire-g++ (--mthumb-interwork is not recognized)? Can I remove -fno-rtti?

→ **MyCalcs** profile · de **Excale** » 30 Mar 2014, 13:00

parisse a écrit:I have set the emulator with the 3.1 os and ndless, and I can start a debugging session, but as soon as I try to debug in the parser code, the system reboots. More precisely when I enter s in the gdb console for stepping in giac_yyparse, then it reboots. I suspect that the memory is almost full, and a memory allocation fails at the begin of the parser code.

I cannot start a debugging session on OS 3.1 when it is running. It always crashes the emulator.
So... I start it, wait until the reboot is done then debug what I want to debug.

Also, did you give the large RAM switch to Nspire emu? (/R)

→ **MyCalcs** profile · de **parisse** » 30 Mar 2014, 15:22

Ok, compiling with -Os, giac.tns is now 3.9M, and it runs with 3.6, at least for 1/2+1/3... But I can not set a breakpoint in main, it reboots, perhaps a problem with -Os -g and gdb.
I have updated the giac.tgz archive (without the large gdb file, I have limited net access today).