Re: [ROOT] fgets oddity

From: Jiri Masik (masik@pc203b.fzu.cz)
Date: Wed Jul 18 2001 - 15:57:25 MEST 

Hi,

I suggest the following modification to TUnixSystem.cxx. It should
cause interrupted system calls (by means of signals) to be resumed.
cheers
        Jiri
Jiri Masik <Jiri.Masik@cern.ch> writes: > Hi Fons, > > maybe it is related to the signal handling in Root. Could you explain > what should happen on SIGCHLD? The problem seems to disappear when I > comment SIGCHLD record from the gSignalMap in TUnixSystem.cxx. > > Jiri > > Fons Rademakers <Fons.Rademakers@cern.ch> writes: > > > Hi Jiri, > > > > I've seen the same on my linux machine (run stress two times and the second > > will be scrambled. It is related to the OpenPipe() call. If you could debug > > and fix this for me I would be very happy. > > > > Cheers, Fons. > > > > > > masik@fzu.cz wrote: > > > > > > Hi, > > > > > > has anybody seen a problem reading output of external commands by > > > fgets after popen (gSystem->OpenPipe) call? It can be noticed as a > > > garbage/empty line at the end of test/stress output where system info > > > is being obtained. I've seen this on Debian/PPC and some Debian/x86 > > > (but not all machines fail). As for non-linux boxes I had a chance to > > > test only dxplus(OSF1) at CERN for which it is less frequent but it > > > happens too. Root version is 3.01/06, libc-2.2.3-6, Linux version > > > 2.4.6-pre8, gcc-3.0. > > > > > > Please see the following example > > > > > > -------------------------------------------------------------------------------- > > > Name: poproot2.cxx > > > poproot2.cxx Type: text/x-c++src > > > Description: poproot2.cxx > > > > > > -------------------------------------------------------------------------------- > > > > > > [dxplus02] /afs/cern.ch/user/j/jmasik/public > g++ -I$ROOTSYS/include > > > poproot2.cxx `root-config --libs` > > > [dxplus02] /afs/cern.ch/user/j/jmasik/public > ./a.out > > > fgets returned NULL > > > [dxplus02] /afs/cern.ch/user/j/jmasik/public > ./a.out > > > * OSF1 dxplus02 V4.0 1229 alpha > > > > > > Any information on popen and fgets I have doesn't mention the need to > > > test anything beyond returned pointers. Bad runs have this line in > > > their tracelogs. > > > read(4, 0x30018000, 4096) = ? ERESTARTSYS (To be restarted) > > > > > > Do you have an idea what's wrong? Everything works if I comment out > > > TSystem.h and remove `root-config --libs` from the compile command on > > > Debian/PPC, on OSF removing TSystem.h is enough. > > > > > > Jiri > > > > > > I don't know how to strace on OSF, on Linux it looks like > > > .... > > > open("/data/masik/root-v3.02.gcc3/include/RQ_OBJECT.h", O_RDONLY) = 3 > > > brk(0x103ae000) = 0x103ae000 > > > fstat64(0x3, 0x7fffe938) = 0 > > > mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, > > > 0) = 0x30017000 > > > read(3, "// @(#)root/base:$Name: $:$Id: "..., 4096) = 4096 > > > _llseek(0x3, 0, 0, 0x7fffe8f8, 0) = 0 > > > read(3, "// @(#)root/base:$Name: $:$Id: "..., 4096) = 4096 > > > read(3, "ar *method) \\\n { Emit(\"Chan"..., 4096) = 152 > > > read(3, "", 4096) = 0 > > > pipe([4, 5]) = 0 > > > fork() = 4880 > > > close(5) = 0 > > > fstat64(0x4, 0x7ffff0c8) = 0 > > > mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, > > > 0) = 0x30018000 > > > read(4, 0x30018000, 4096) = ? ERESTARTSYS (To be > > > restarted) > > > --- SIGCHLD (Child exited) --- > > > fstat64(0x1, 0x7fffe9e8) = 0 > > > mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, > > > 0) = 0x30019000 > > > ioctl(1, 0x402c7413, 0x7fffe848) = 0 > > > write(1, "fgets returned NULL\n", 20) = 20 > > > > -- > > Org: CERN, European Laboratory for Particle Physics. > > Mail: 1211 Geneve 23, Switzerland > > E-Mail: Fons.Rademakers@cern.ch Phone: +41 22 7679248 > > WWW: http://root.cern.ch/~rdm/ Fax: +41 22 7677910

This archive was generated by hypermail 2b29 : Tue Jan 01 2002 - 17:50:52 MET