Warn the user that there's something wrong.

Okay. But "Something's wrong!" isn't very useful to the user.

According to POSIX, close can fail with one of 3 errors:

1. EBADF: fd isn't a valid open file descriptor.

   Hopefully, if this can occur in the application, it'll be detected/sorted before the user gets his hands in it.

2. EINTR: The close() call was interrupted by a signal; see signal(7).

   Ditto. If this is a possibility, a signal handler will have been installed to handle it.

3. EIO: An I/O error occurred.

   And we're back to "Something's wrong!".

And then, the man page says:"A successful close does not guarantee that the data has been successfully saved to disk, as the kernel defers writes.", so even a successful close doesn't guarantee good data.

So we've told the user, "Something's wrong!, but not what. And even if we haven't told him, something could still be wrong. So where does that leave him?

He could decide to re-run the process to recreate the file; but what if it (whatever "it" is) happens again?

And, it could be he does so unnecessarily, because it was only the close that failed and not the preceding writes.

And what can he do about the failures that happen after we've closed the file; due to caching?

In the end, the only arbiter of whether the file is good is whether the next use of that file is successful.

So, the only sure way to detect the types of problems that might be indicated by close failing, is to ensure that the next process in the chain; even if that is a human being reading the file, can adequately detect that the file is incomplete or otherwise corrupted. Perhaps a sentinel record ("The end") or an known file size.

And whatever mechanism is put in place for the next process to detect the error; negates any purpose in detecting and warning that "Something's wrong!";

Did I miss a possibility?

Perl close() is not POSIX close(). Some less-bizarre reasons that Perl close() might fail include:

ENOSPC

A previous write to disk failed because the disk partition ran out of space (and you had not cleared errors for that handle by seek()ing since then).

EPIPE

You had unflushed data and close() tried to flush it and all of the readers at the other end of your pipe/socket have already closed/shutdown their end. Or a previous write to that handle failed for that reason (and you probably got a SIGPIPE earlier).

EIO

Your device has decided to fail. This is more specific than "something is wrong". You should arrange to remember what you had opened so that you can mention that in the "close() failed" error message so the user (or their admin) can determine which device to check.

ECONNRESET

Your socket connection got reset (since the last time you cleared errors by seek()ing).

ENOMEM

The system wanted to allocate more buffer space but the call to allocate that failed (possible at least for sockets).

Plus there are errors that can originate from some I/O layer. You'd have to consult the documentation and/or code for whatever layers you might use for more information on that. Perhaps you can use an encoding layer configured to complain about untranslatable characters via an error return. Or a "decompress" layer might complain about invalid input.

It can be useful to check for close failing because it may be important to not continue on after close() failed because the data you were writing or reading is incomplete and so you shouldn't submit it or shouldn't mark the data source as "done" or whatever.

It is usually easier to check for close() failing than to check every single I/O operation for failure. If using buffered I/O, then checking every single I/O operation wouldn't be sufficient anyway.

If I have no logical recovery step for "incomplete data", then I will just warn when close fails.

- tye        

If I have no logical recovery step for "incomplete data", then I will just warn when close fails.

And that is really the point I'm arriving at; and your clarification of some of possibilities has taken me ever closer.

Basically, there is never a logical recovery step; because the actually failure that gets reported by close may have happened hours ago; and because the range of possibilities is so diverse; and because it would be literally impossible to program a recovery for the vast majority of them.

So, (IMO) the only logical step would be to modify close so that it automatically logged any errors.

It ought be able to extract filenames from handles, ips/socknums from sockets etc. and generally produce a much more detailed account of the failure than user code could hope to do.

And it would relieve the programmer from that burden. I guess I would like an autowarn rather than autodie pragma.

---

With the rise and rise of 'Social' network sites: 'Computers are making people easier to use everyday'

Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.

"Science is about questioning the status quo. Questioning authority".

In the absence of evidence, opinion is indistinguishable from prejudice.

I'm with torvalds on this Agile (and TDD) debunked I told'em LLVM was the way to go. But did they listen!

Okay. But "Something's wrong!" isn't very useful to the user.

close $fh
  or croak("E: Unable to close file handle for $file: $!");
[download]

This usually results in a nice explanation what went wrong.

And, it could be he does so unnecessarily, because it was only the close that failed and not the preceding writes.

This also catched me by surprise. The 'print {$fh}' did return a success, it was only apparent when closing the file handle, that Perl was actually unable to write.

And what can he do about the failures that happen after we've closed the file; due to caching?

Out of scope. As least as far as Perl is concerned. (And I think there is no chance you can truly validate that the data has been written to disk. At least as long as you do not have an exact knowledge about the storage architecture. And even then it may still be lying.)

croak("E: Unable to close file handle for $file: $!"); This usually results in a nice explanation what went wrong.

Do you have, or can you point me at a list of the possibilities? Or even a few of them?

---

With the rise and rise of 'Social' network sites: 'Computers are making people easier to use everyday'

Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.

"Science is about questioning the status quo. Questioning authority".

In the absence of evidence, opinion is indistinguishable from prejudice.

I'm with torvalds on this Agile (and TDD) debunked I told'em LLVM was the way to go. But did they listen!