search

dstndstn / astrometry.net

Astrometry.net -- automatic recognition of astronomical images
http://astrometry.net
Other
679 stars 189 forks source link

'make clean' doesn't clean everything #161

Closed DerPit closed 2 years ago

DerPit commented 5 years ago

Hi,

I've so far been using astrometry.net (version 0.78) on my laptop (i7-6820HQ) without issues. I recently got myself a mini-PC (ASUS PN40, Celeron N4000) with the same OS and version installed as on the laptop (openSUSE Tumbleweed, latest version), intended as mount/equipment control for my telescope, and wanted to check if it is good enough to solve by itself.

So I had just copied over the installation. However, trying to run solve-field (even without options) immediately lead to a coredump due to SIGILL (ILL_ILLOPN). First time I have such issues, I compile all my stuff for the mini on my laptop. So I installed all needed dependencies and compiled on the mini-PC. Now solve-field w/o options does display the help. But running the demo 'solve-field --scale-low 10 demo/apod4.jpg' again lead to a coredump with SIGILL.

I'm not really a programmer, but tried to run this in gdb. Here's the output: `Program received signal SIGILL, Illegal instruction. 0x000000000044b4bf in log_init_structure (logger=0x50fd00 , level=level@entry=LOG_MSG) at log.c:43 43 logger->t0 = timenow(); (gdb) bt

0 0x000000000044b4bf in log_init_structure (logger=0x50fd00 ,

level=level@entry=LOG_MSG) at log.c:43

1 0x000000000044b509 in log_init (level=level@entry=LOG_MSG) at log.c:49

2 0x0000000000404098 in main (argc=4, args=0x7fffffffe098)

at solve-field.c:933

` I'm at a loss now. Any suggestions what to look for? Am I missing something? Has that to do with the processor?

dstndstn commented 5 years ago

Hi,

You compiled the code on the mini-PC? Normally, we use the "-march=native" compiler flags to automatically choose compile options and CPU features that are available on the compiling machine. So assuming you did a clean compile on the mini-PC, it should have made programs it could run...

You can turn off that by setting the ARCH_FLAGS environment variable before running "make". (You'll have to do a "make clean" to remove the existing .o and executable files. eg,

make clean ARCH_FLAGS= make

Thanks for the bug report. I hope this works!

DerPit commented 5 years ago

Thanks so much for this rapid feedback. Yes, I had assumed that there might be machine-specific compile options, and had run a 'make reconfig' before the make. Obviously that wasn't enough - a 'find . -name *.o -ls' revealed a whole bunch of compiled files from the old system. So I ran make clean now(*) and recompiled/installed everything. Now it runs as expected. Jippieh!

Thanks so much for your help!

(*) after the make clean there were still a few files left: ./sdss/cutils_wrap.o ./sdss/_cutils.so ./blind/_plotstuff_c.so ./util/_util.so ./libkd/spherematch_c.so I had also removed those myself, to be on the safe side :)

dstndstn commented 2 years ago

fixed in 871364986cd938a4eab22de499ba12734d6091ad