burzio@mmlai.UUCP (Tony Burzio) (11/24/89)
System: HP-UX 6.5 B 9000/370 While trying out the C DIRECTORY(3C) routines, I noticed some very strange behavior. When you do a readdir, the files are returned in what seems to be random order. At first it appeared that they return in the order that the files are created, but new files pop up where ever. The DIR pointer is not closed until NULL is returned from readdir. Has anybody else used this code and been able to extract file names in alphabetical order? Where should I look for more info besides the HP-UX reference? On a related note, is it possible to run a sort on a tmpfile() created file? I really need the filenames in order... Thanks in advance. ********************************************************************* Tony Burzio * SNOW! On Thanksgiving? Ski time! Martin Marietta Labs * mmlai!burzio@uunet.uu.net * *********************************************************************
donn@hpfcdc.HP.COM (Donn Terry) (11/25/89)
(Almost?) all UNIX systems, not just HP-UX, return the directory enties in
unspecified order. (The actual order is a function of the order
in which the entries were created (and removed) and of the length of
the entries if you are using long file names.)
The simplest (although not the most efficient in terms of system
resources) solution to get what you want would be
foo = popen("ls", "r");
and do freads from foo to get the names.
You could read them all with readdir() and use qsort(), or copy them
to an ordinary file and use the sort command via system().
Donn Terry
HP Ft. Collins
(No, I'm not speaking officially for HP, or anyone else for that matter.)bae@hp-lsd.COS.HP.COM (Bruce Erickson) (11/26/89)
>While trying out the C DIRECTORY(3C) routines, I noticed >some very strange behavior. When you do a readdir, >the files are returned in what seems to be random >order. At first it appeared that they return in the >order that the files are created, but new files pop >up where ever. I'm not an expert, but I believe that 'readdir' returns the files in the order that they are in the directory inode, which I think is essentially an array of file names, which are placed in the first unused slot. So, when a directory is first created the files are in first-come first-slot basis; after a file is removed, however, I believe that the next created file gets the slot freed up by the removed file. (I am remembering this from an O/S class 8 years ago, so I may be mis- remembering this!) If you need the files in alphabetical order, read them in but stick them in a binary tree (use bsearch(3C) for ease of implementation) then read them back from the binary tree.... Bruce Erickson hp-lsd!bae
markf@hpupnja.HP.COM (Mark Fresolone) (11/26/89)
The 'ls' command has us all spoiled! It turns out that directory entries are not actually stored in any order. The directory(3) routines simply report 'em as they see 'em. Entries are kept rather primitively in a static array. At the beginning of a directory's life, files appear in creation order. However, when a file or directory is "removed", its entry is simply marked unused (traditionally, inode == 0). The next file created fills the first available unused slot, allocating a new block of entries when necessary. To visualize the behavior, create a directory, and then create and delete files, using "strings ." or "od -c ." between operations. Good coding! #include <disclaimer.h> Mark Fresolone hplabs!hpfcse!hpupnja!markf
fkittred@bbn.com (Fletcher Kittredge) (11/26/89)
In article <616@mmlai.UUCP> burzio@mmlai.UUCP (Tony Burzio) writes: >System: HP-UX 6.5 B 9000/370 > >While trying out the C DIRECTORY(3C) routines, I noticed >some very strange behavior. When you do a readdir, >the files are returned in what seems to be random >order. At first it appeared that they return in the >order that the files are created, but new files pop >up where ever. The DIR pointer is not closed until >NULL is returned from readdir. Has anybody else used >this code and been able to extract file names in >alphabetical order? Where should I look for more info >besides the HP-UX reference? On a related note, is it >possible to run a sort on a tmpfile() created file? I >really need the filenames in order... Thanks in advance. Tony; If you intend to do programming on a Unix system, I strongly recommend you buy "The Unix Programming Environment" (Kernighan and Pike, 1984), and "The Design of the Unix Operating System" (Bach). The basic answer to your question is that you are assuming that directories are stored in sorted order. They are not; there is no reason for them to be stored sorted. You can look at the order they are stored in by entering the command: % strings <directory> |more For a justification of this design decision, any good intro to algorithms would help; my favorites are "Data Structures and Algorithms" (Aho, Hopcroft and Ullman, 1983), and Sedgewick's "Algorithms" (1988?). These would also be good places to look for sort algorithms to get your file names in order! happy hunting, fletcher P.S. If I read the manual correctly, you won't be able to sort a file created by tmpfile(), since it disappears from the file system immediately after it created. Fletcher E. Kittredge fkittred@bbn.com
rer@hpfcdc.HP.COM (Rob Robason) (11/30/89)
> System: HP-UX 6.5 B 9000/370 > Has anybody else used this code and been able to extract file names in > alphabetical order? The 6.5 release has an undocumented function in libc called scandir and another called alphasort. These derive from BSD and will read an entire directory and sort it alphabetically. Note that use of scandir and alphasort is not yet supported on the 800, not for any particular reason -- just that it got into the 300 more by accident than plan. Attached is the reference page for scandir. HP hasn't tested the scandir function in 6.5, and since it's not documented it is officially "unsupported", but the sample application shown at the bottom of this response does seem to work. > I really need the filenames in order... Thanks in advance. > Tony Burzio * SNOW! On Thanksgiving? Ski time! Rob Robason * Went skiing yesterday at Keystone -- GREAT!! Not speaking for HP, just trying to be helpful ######################################################################## SCANDIR(3) (September 17, 1985) SCANDIR(3) NAME scandir, alphasort - scan a directory SYNOPSIS #include <sys/types.h> #include <sys/dir.h> scandir(dirname, namelist, select, compar) char *dirname; struct direct *(*namelist[]); int (*select)(); int (*compar)(); alphasort(d1, d2) struct direct **d1, **d2; DESCRIPTION Scandir reads the directory dirname and builds an array of pointers to directory entries using malloc(3). It returns the number of entries in the array and a pointer to the array through namelist. The select parameter is a pointer to a user supplied subroutine which is called by scandir to select which entries are to be included in the array. The select routine is passed a pointer to a directory entry and should return a non-zero value if the directory entry is to be included in the array. If select is null, then all the directory entries will be included. The compar parameter is a pointer to a user supplied subroutine which is passed to qsort(3) to sort the completed array. If this pointer is null, the array is not sorted. Alphasort is a routine which can be used for the compar parameter to sort the array alphabetically. The memory allocated for the array can be deallocated with free (see malloc(3)) by freeing each pointer in the array and the array itself. SEE ALSO directory(3), malloc(3), qsort(3), dir(5) DIAGNOSTICS Returns -1 if the directory cannot be opened for reading or if malloc(3) cannot allocate enough memory to hold all the data structures. Hewlett-Packard Company - 1 - Nov 29, 1989 ########################################################################## #include <sys/types.h> #include <sys/dir.h> int alphasort(); main() { struct direct **namelist; scandir(".", &namelist, 0, alphasort); while (*namelist) { printf("%s\n", (*namelist)->d_name); namelist++; } }