[ http://issues.apache.org/jira/browse/STDCXX-308?page=comments#action_12457393 ] Martin Sebor commented on STDCXX-308: -------------------------------------
-------- Original Message -------- Subject: Re: aCC 3.70 -AA fd's close not being called. Date: Fri, 08 Dec 2006 11:47:51 -0700 From: Martin Sebor <[EMAIL PROTECTED]> Organization: Rogue Wave Software To: Dennis Handly <[EMAIL PROTECTED]> CC: [EMAIL PROTECTED], [EMAIL PROTECTED] References: <[EMAIL PROTECTED]> Dennis Handly wrote: > It appears this problem is still in the apache code: > http://svn.apache.org/repos/asf/incubator/stdcxx/trunk/include/fstream.cc > > It seems to violate 27.8.1.3(6), which says it always "finally" calls > "as if" fclose(3). And returns an error if any calls fail. I agree that it's a bug that the dtor leaks the fd when close() fails. The question is whose fault is it, ours or the standard's? ;-) I don't think the text is clear about close() being required to call fclose() even if overflow() fails (or throws). Further, suppose close() is being called directly (as opposed from the dtor). Should it close the fd even if overflow() failed due to some transient and recoverable error? Since close() indicates success or error to the caller I think it should only call fclose() if overflow() returns successfully. Otherwise it should leave it open so that the caller can fix the problem and not lose data. OTOH, the dtor should never leak the fd and should be required to close it regardless of whether overflow is or isn't successful, and it should probably not propagate exceptions. But to do that the spec needs to specify this behavior explicitly. Let me raise it on the reflector. Martin > filebuf dtor doesn't close fd after exception > --------------------------------------------- > > Key: STDCXX-308 > URL: http://issues.apache.org/jira/browse/STDCXX-308 > Project: C++ Standard Library > Issue Type: Bug > Components: 27. Input/Output > Affects Versions: 4.1.2, 4.1.3 > Environment: all > Reporter: Martin Sebor > Assigned To: Martin Sebor > > -------- Original Message -------- > Subject: Re: aCC 3.70 -AA fd's close not being called. > Date: Thu, 7 Dec 2006 23:17:55 -0800 (PST) > From: Dennis Handly <[EMAIL PROTECTED]> > To: [EMAIL PROTECTED] > CC: [EMAIL PROTECTED], [EMAIL PROTECTED], [EMAIL PROTECTED] > It appears this problem is still in the apache code: > http://svn.apache.org/repos/asf/incubator/stdcxx/trunk/include/fstream.cc > It seems to violate 27.8.1.3(6), which says it always "finally" calls > "as if" fclose(3). And returns an error if any calls fail. > ================================================================ > >From: "Levie, Barry" <[EMAIL PROTECTED]> > Date: Thu, 7 Dec 2006 15:36:33 -0000 > Subject: aCC 3.70 -AA fd's close not being called. > I am not sure if this exactly represents the problem the customer is > having as I am still waiting for more details (versions/test case etc > etc). However this could be close and I think it shows a problem. > Why does the ofstream destructor not close the file descriptor when its > called after its thrown an exception for "out of disk space". > This aCC 3.70 -AA example shows an ever increasing number of open file > descriptors in glance > # what /usr/lib/libstd_v2.2: > HP aC++ B3910B A.03.30 C++ Standard Library (RogueWave Version 2.02.01) > #include <iostream> > #include <fstream> > #include <string.h> > using namespace std; > struct dummy { > dummy() { cout << "\ndummy created \n" ;} > ~dummy() { cout << "\ndummy deleted \n"; } > }; > void bar() { > char buf[1024*1024]; > memset(buf,'A',1024*1024); > buf[1024*1024 -1] = 0; > ofstream of; > dummy df; > of.exceptions(ofstream::eofbit | ofstream::failbit | ofstream::badbit); > try { > of.open("/home/test.txt"); // this is on a volume that is nearly full. > while(1) { > of << buf; > } > } > catch(ofstream::failure& e) { > cout << "\nException write/file"; > } > } > int main () { > while(1) bar(); > } > # ./a.out > dummy created > msgcnt 160 vxfs: mesg 001: vx_nospace - /dev/vg00/lvol5 file system full > (1 block extent) > Exception write/file > dummy deleted > .... > Glance output shows > D File Name Type Mode Count Offset > --------------------------------------------------- > 0 /dev/pts/ta stream rd/wr 6 333822 > 1 /dev/pts/ta stream rd/wr 6 333822 > 2 /dev/pts/ta stream rd/wr 6 333822 > 3 /home/test.txt reg write 1 101957632 > 4 /home/test.txt reg write 1 101957632 > 5 /home/test.txt reg write 1 101957632 > ... > ================================================================ > >From: Dennis Handly <[EMAIL PROTECTED]> > Date: Thu, 7 Dec 2006 23:03:08 -0800 (PST) > To: [EMAIL PROTECTED], [EMAIL PROTECTED] > Subject: Re: aCC 3.70 -AA fd's close not being called. > >From: "Levie, Barry" <[EMAIL PROTECTED]> > >I am not sure if this exactly represents the problem the customer is > >having as I am still waiting for more details (versions/test case etc > >etc). However this could be close and I think it shows a problem. > >Why does the ofstream destructor not close the file descriptor when its > >called after its thrown an exception for "out of disk space". > A highly modified version works for me. > And it isn't closed on the throw but by falling off the bar body. > >This aCC 3.70 -AA example shows an ever increasing number of open file > >descriptors in glance > Hmm. > ># what /usr/lib/libstd_v2.2: > > HP aC++ B3910B A.03.30 C++ Standard Library (RogueWave Version 2.02.01) > This obsolete version isn't supported. > Here is what I have and it shows close being called after I throw. > I've manually modified the write return to -1 and that throws and is caught. > (The debugger was broken and it didn't really happen. )-: > #include <iostream> > #include <fstream> > #include <string.h> > #include <stdio.h> > using namespace std; > struct dummy { > dummy() { fprintf(stdout, "dummy created\n"); fflush(stdout); } > ~dummy() { fprintf(stdout, "dummy deleted\n"); fflush(stdout); } > }; > extern "C" int _close(int fd); > extern "C" int close(int fd) { > if (fd > STDERR_FILENO) { > fprintf(stdout, "close(%d)\n", fd); > fflush(stdout); > } > return _close(fd); > } > char buf[1024]; > void bar() { > ofstream of; > dummy df; > of.exceptions(ofstream::eofbit | ofstream::failbit | ofstream::badbit); > try { > of.open("test.txt"); // this is on a volume that is nearly full. > fprintf(stdout, "FD = %d\n", of.rdbuf()->fd()); > fflush(stdout); > for (int i = 0; i < 2; ++i) > of << buf; > throw ios_base::failure(); > } > catch(ofstream::failure &e) { > fprintf(stdout, "Exception write/file\n"); > fflush(stdout); > } > } > int main() { > memset(buf,'A',sizeof(buf)); > buf[sizeof(buf)-1] = '\0'; > int i; > for (i = 0; i < 10; ++i) > bar(); > } > Ah, it's catch 22. When it destroys ofstream it calls close, which calls > overflow which calls _C_write which calls write which gets an error and > then returns before closing. > It seems it is working as designed be RW. > But this seems to violate 27.8.1.3(6), which says it always calls > fclose(3). > File a CR. -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa - For more information on JIRA, see: http://www.atlassian.com/software/jira
