[fpc-devel]Re: Bug in AssignPipe() on Linux

[Marco van de Voort]

[ Charset ISO-8859-1 unsupported, converting... ]
> Marco van de Voort wrote:
> > - Do an strace, and see if there is a function that fails.
> 
> Everything seems OK.
> 
> $ strace pipetest
> execve("/home/plc/Pascal/test1.9/pipetest", ["pipetest"], [/* 63 vars */]) = 0
> sigaction(SIGFPE, {0x805a9d4, [], 0}, {SIG_DFL}, 0x40054d58) = 0
> sigaction(SIGSEGV, {0x805a9d4, [], 0}, {SIG_DFL}, 0x40054d58) = 0
> sigaction(SIGBUS, {0x805a9d4, [], 0}, {SIG_DFL}, 0x40054d58) = 0
> sigaction(SIGILL, {0x805a9d4, [], 0}, {SIG_DFL}, 0x40054d58) = 0
> getpid()                                = 4307
> readlink("/proc/4307/exe", "/home/plc/Pascal/test1.9/pipetest", 255) = 33
> open("/etc/timezone", O_RDONLY)         = 3

This is the culprit. People that have this file don't have the problem.
If this one goes wrong, the errno value remains there, and the next 
decision that bases itself on linuxerror goes wrong. 

This is why it was hard to reproduce:
 
In my case, this happens

getpid()                                = 3953
readlink("/proc/3953/exe", "/export/home/marcov/pipetest", 255) = 28
open("/etc/timezone", O_RDONLY)         = -1 ENOENT (No such file or directory)

-> failed 

stat("/etc/localtime", {st_mode=S_IFREG|0644, st_size=1074, ...}) = 0
open("/etc/localtime", O_RDONLY)        = 3

-> found, but errno is untouched.