88341 - [RFE] support for file truncation (ftruncate, etc.)

Status: NEW

(NSPR :: NSPR, enhancement, P3)

Description

[Darin Fisher]

This request results from mozilla bug 86474, in which file truncation is needed
by the necko cache.  Since the truncation system call varies across platforms this
is a definite candidate for inclusion in NSPR.

Under BSD/XOPEN_EXTENDED compliant unixen there is the ftruncate system call.
Under Win32 there is the SetEndOfFile system call.  And, under the MacOS there
is the SetEOF system call.  I have no idea what equivalent call BeOS or OS/2 
provide.

Comment 1

[timeless]

beos...
#include <unistd.h>
extern int      ftruncate(int fd, off_t newsize);

Comment 2

[Mike Kaply [:mkaply]]

On OS/2, there is API DosSetFileSize:

ulrc = DosSetFileSize(hFile, cbSize);

Comment 3

[Wan-Teh Chang]

Larry, I suggest that we get this done in NSPR 4.2.

Use the PROffset64 type for the new file size, i.e.,
only add the 64-bit version of this I/O method.

PRStatus PR_Truncate(PRFileDesc *fd, PROffset64 newsize);

or PR_SetEndOfFile, or PR_SetFileSize.

Comment 4

[Wan-Teh Chang]

Adding a new function requires bumping the minor version number.
Since this enhancement is requested by the Mozilla client team,
we need to land NSPR 4.2 Beta on the NSPR client tag/branch
first (bug 78471).

Comment 5

[larryh (gone)]

Reassign to wtc.

Comment 6

[Darin Fisher]

FWIW: I would prefer PR_SetEndOfFile.  It would rely on a call to PR_Seek before
truncating at a specific offset, but that is really not a problem since
truncation at a specific offset is typically used before writing at that offset
(as is the case in mozilla/necko).

Consider the scenario on UNIX w/ PR_Truncate

   PR_Seek(offset);
   PR_Truncate(offset);
       ftruncate(offset);

and on WIN32 w/ PR_Truncate

   PR_Seek(offset);
   PR_Truncate(offset);
       prev = PR_Tell();
       PR_Seek(offset);
       SetEndOfFile();
       PR_Seek(prev);  // restore file pointer to previous value

now, consider PR_SetEndOfFile on UNIX
  
   PR_Seek(offset);
   PR_SetEndOfFile();
       ftruncate(PR_Tell());

and, PR_SetEndOfFile on WIN32

   PR_Seek(offset);
   PR_SetEndOfFile();
       SetEndOfFile(); 
       
using PR_SetEndOfFile definitely appears to result in fewer system calls on
WIN32 in this common usage scenario.

anyways, that's just my $.02

Comment 7

[Wan-Teh Chang]

Because of Larry's departure, I won't be able to implement
this new function.  If anyone volunteers to take this bug,
I will be happy to answer questions and review patches.

Comment 8

[Wan-Teh Chang]

Darin, it seems that you are making a distinction between
PR_Truncate and PR_SetEndOfFile.  I named them as two
possible names of the new function (to set file size and
extend or truncate as appropriate).  I don't understand
your code samples and what your points are.  I must
have missed the context for your comments.  Sorry.  Could
you explain?

Comment 9

[Darin Fisher]

my examples are meant to demonstrate what code would be run if NSPR exported a
PR_Truncate method or a PR_SetEndOfFile method for a common usage senario,
namely "truncate an open file to a specific offset and then start writing at
that offset."

in this usage senario, client code will naturally call PR_Seek before (or after)
either PR_Truncate or PR_SetEndOfFile, since that sets the write cursor.  so,
for each platform we should consider what the underlying code looks like to help
us decide which function should be exported from NSPR.

in my examples (above) i've added extra indentation to the code NSPR would use
to implement PR_Truncate and PR_SetEndOfFile, and i think this shows that
PR_SetEndOfFile would be a more practical API (at least in this particular usage
senario).

anyways, i'd be up for cutting a patch once we decide on a good API.

Comment 10

[gordon]

I'll see if I can work up a patch next week.

Comment 11

[Brant Gurganus]

[RFE] is deprecated in favor of severity: enhancement.  They have the same meaning.

Comment 12

[gordon]

Additional configuration suggestions in bug 162588.

Comment 13

[gordon]

Attachment: Truncate() currently used by disk cache

This attachment is the listing of the Truncate() method as it's currently
implemented in the disk cache.

Comment 14

[gordon]

Attachment: proposed NSPR version of PR_Truncate()

This attachment is an initial proposal for an NSPR version of the Truncate()
call.  It hasn't been tested, and doesn't take into account I/O layers.

I'm proposing a variation on ftruncate() because that's what we're currently
using in the tree, and it doesn't require a seek() beforehand.

Comment 15

[gordon]

Reassigning bug as per conversation with Wan-Teh.

Comment 16

[BugBot [:suhaib / :marco/ :calixte]]

The bug assignee didn't login in Bugzilla in the last 7 months, so the assignee is being reset.