Re: [RFC/PATCH] Modifying internal FS transaction properties

2015-03-03 Thread Evgeny Kotkov
Evgeny Kotkov writes: > Okay, here is what I am about to do: > > 1) Exploit the BDB problem with SVN_FS_TXN_CLIENT_DATE not working as >expected in a separate test. Probably, I am also going to create an >issue, because without a fix we'd be shipping Subversion 1.9 with a >new transa

Re: [RFC/PATCH] Modifying internal FS transaction properties

2015-03-03 Thread Philip Martin
Evgeny Kotkov writes: > 1) Exploit the BDB problem with SVN_FS_TXN_CLIENT_DATE not working as > expected in a separate test. Probably, I am also going to create an > issue, because without a fix we'd be shipping Subversion 1.9 with a > new transaction flag that doesn't work properly on every FS

Re: [RFC/PATCH] Modifying internal FS transaction properties

2015-03-03 Thread Evgeny Kotkov
Julian Foad writes: >> Using svn:date to store txn creation date doesnt't work so well now that >> we allow clients to set svn:date. It might have been better if the txn >> creation date was stored in a separate txn property that was deleted on >> commit. If we were to do that then we would nee

Re: [RFC/PATCH] Modifying internal FS transaction properties

2015-03-03 Thread Julian Foad
Philip Martin wrote: > [...]  Both BDB and FSFS set the date on the txn when the txn is > created.  This was introduced so that admins could look at the date and > get the txn creation date which is useful when deciding whether or not > to delete old txns.  The BDB bug is that this setting should

Re: [RFC/PATCH] Modifying internal FS transaction properties

2015-03-03 Thread Philip Martin
Evgeny Kotkov writes: >I propose that we solve this part of the problem with the attached patch. >Please note that while preparing this patch, I also unveiled that the >SVN_FS_TXN_CLIENT_DATE flag might not work as expected with BDB. >I think we should fix this with a separate pat

Re: [RFC/PATCH] Modifying internal FS transaction properties

2015-03-03 Thread Julian Foad
Evgeny Kotkov wrote: > 1) Currently we allow changing 'svn:check-ood' and 'svn:check-locks'.  We also >    silently discard a property change whenever it targets 'svn:client-date'. >    I find the first part questionable, because we basically allow a caller to >    mess with our private implementat

[RFC/PATCH] Modifying internal FS transaction properties

2015-03-02 Thread Evgeny Kotkov
I'd like to draw attention to one aspect of the FS API that received a change in Subversion 1.9. We have 2 functions, svn_fs_change_txn_prop() and svn_fs_change_txn_props(), that allow changing properties of the existing transactions in a filesystem. I think we might want to revisit their behavior