[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: svn commit: r1073366 - in /subversion/trunk: notes/wc-ng/pristine-store subversion/libsvn_wc/wc-queries.sql subversion/libsvn_wc/wc_db_pristine.c

From: Branko Čibej <brane_at_e-reka.si>
Date: Sat, 26 Feb 2011 10:35:47 +0100

On 26.02.2011 07:32, Ivan Zhakov wrote:
> 2011/2/26 Branko Čibej <brane_at_e-reka.si>:
>> On 25.02.2011 16:53, Julian Foad wrote:
>>> On Thu, 2011-02-24, Branko Čibej wrote:
>>>> On 24.02.2011 18:03, Julian Foad wrote:
>>>>> On Wed, 2011-02-23, Daniel Shahaf wrote:
>>>>>> julianfoad_at_apache.org wrote on Tue, Feb 22, 2011 at 15:38:35 -0000:
> [...]
>
>>>> It is not the business of the wc_db+pristine-store to track
>>>> every process that happens to have an open handle to the pristine file.
>>>> A deletion of the pristine file should succeed even if there are open
>>>> handles referring to it.
>>> So you're suggesting we should promise that a reader can continue
>>> reading the file (at least once through to the end, not sure about
>>> rewinding) even if something else deletes the file from the store part
>>> way through. I think you're suggesting those semantics are more
>>> reasonable than "you have to hold some sort of lock while you read it",
>>> which is what my design boiled down to.
>> Yes, indeed, they're far more reasonable because the OS already gives
>> them to you. On Unix, when you delete a file, it vanishes from the
>> directory; but open handles remain valid, and the backing store of the
>> data still exists. The file only really goes away when the last handle
>> is closed.
>>
>> On Windows, the situation is pretty much the same (assuming
>> FILE_SHARE_DELETE which we've already determined APR always does --
>> guess why :), *except* that the file only vanishes from the directory
>> after it's been deleted once the last handle to it is closed, that's why
>> I mentioned the tricky part of re-instating the file.
>>
> [..]
>>> I guess I'll have to figure out how to implement this "trifle more
>>> involved" part on Windows, now.
>> Lucky you, the name of the file is the digest of its contents, so in
>> order to reinstate the file on Windows you only have get the system to
>> twiddle it's "deleted" bit. "Only." I seem to recall that's not even
>> hard to do, but my last battle with Windows filesystem internals was
>> more than 10 years ago. If you can't find relevant docs, you could try
>> asking APR for that functionality. I'm sure Will Rowe will give you a
>> dozen reasons why doing that is not a good idea, and also explain how to
>> do it. :)
>>
> Problem of re-installing file over marked for deletion file can be
> solved using the following trick:
> 1. Rename file to temporary name.
> 2. Delete it

(If the proper share bits are set.)

Yes, that'd work, but if there's a way to unmark the deletion bit,
that's even better, since then you'd not even have to create another
file (with identical contents).

-- Brane
Received on 2011-02-26 10:36:31 CET

This is an archived mail posted to the Subversion Dev mailing list.

This site is subject to the Apache Privacy Policy and the Apache Public Forum Archive Policy.