Bug 21276
| Summary: | libI77 is unable to handle files >2Gbytes | ||||||
|---|---|---|---|---|---|---|---|
| Product: | Base System | Reporter: | vazquez <vazquez> | ||||
| Component: | gnu | Assignee: | Alexander Kabaev <kan> | ||||
| Status: | Closed FIXED | ||||||
| Severity: | Affects Only Me | ||||||
| Priority: | Normal | ||||||
| Version: | 4.1-STABLE | ||||||
| Hardware: | Any | ||||||
| OS: | Any | ||||||
| Attachments: |
|
||||||
The following patch allows random direct access files bigger than >2Gbytes on FreeBSD/x86, it applies against the netlib libI77 sources. To integrate this on gcc would require that autoconf tests all the 32 bit target archs for: 1) The presence of fseeko/ftello or fseek64/ftell64 or... the appropriate pair of 64bit off_set functions; 2) What argument these functions use: off_t, 'long long'; 3) Other targets (16bit, 64bit, linux/x86) should be left untouched (that is, it seems this patch is specific to *BSD intel) an alternative way would be to submit these patchs to David Gay at Bell (the f2c/lib*77 mantainer) and hope this will be imported on gcc sometime and from there into FreeBSD tree. Pedro This bug was apparently introduced with the import of contrib/f2c. Recently the gcc 3.1 version of f2c was imported into FreeBSD 5.0-CURRENT; this version appears to address at least part of this issue with the use of off_t instead of long. Cheers, Mark. -- Mark Valentine, Thuvia Labs <mark@thuvia.co.uk> <http://www.thuvia.co.uk> "Tigers will do ANYTHING for a tuna fish sandwich." Mark Valentine uses "We're kind of stupid that way." *munch* *munch* and endorses FreeBSD -- <http://www.calvinandhobbes.com> <http://www.freebsd.org> Responsible Changed From-To: freebsd-bugs->kan Assign to gcc maintainer State Changed From-To: open->closed Fixed in GCC 3.x series. GCC 2.95.x is closed for development. |
libI77 is unable to handle files greater than 2147627376 bytes (2G), this makes the g77 (and f2c/gcc too on the latest freebsd versions ) generated binaries unable to deal with large quantum chemistry calculations, for example Fix: I'm not sure but it seems this bug may be related to this modification: /usr/src/contrib/libf2c/libI77/Version.c /* 19 Jan. 1998: backspace.c: for b->ufmt==0, change sizeof(int) to sizeof(uiolen). On machines where this would make a difference, it is best for portability to compile libI77 with -DUIOLEN_int (which will render the change invisible). */ According to /usr/src/contrib/libf2c/libI77/README.netlib Unformatted sequential records consist of a length of record contents, the record contents themselves, and the length of record contents again (for backspace). Prior to 17 Oct. 1991, the length was of type int; now it is of type long, but you can change it back to int by inserting #define UIOLEN_int at the beginning of fio.h. This affects only sue.c and uio.c . I'm not an expert on the recent Make buildworld structure to follow all steps from the import of libf2c into egcs to trace this problem but I believe somewhere from 1996 to now something was changed and limited the IO routines to 2G. How-To-Repeat: Try to create a file bigger than 2G using fortran. Run any big quantum chemistry ab-initio calculation , I've old binaries (from 1996) of gaussian (www.gaussian.com) compiled on FreeBSD2.x that run fine with big files, I found this when building new binaries with FreeBSD4.x