[PATCH 2 of 6] transaction: use unfiltered changelog length for transaction start

[Pierre-Yves David]

On 05/24/2017 11:17 PM, Martin von Zweigbergk wrote:
> On Fri, May 19, 2017 at 1:05 PM, Pierre-Yves David
> <pierre-yves.david at ens-lyon.org> wrote:
>>
>>
>> On 05/18/2017 08:23 PM, Durham Goode wrote:
>>>
>>> # HG changeset patch
>>> # User Durham Goode <durham at fb.com>
>>> # Date 1495129620 25200
>>> #      Thu May 18 10:47:00 2017 -0700
>>> # Node ID 35235b1cefb101dad0672a4c4ee9486fba8b935b
>>> # Parent  19be454ee16925d01da8f9d329cb48556083da1b
>>> transaction: use unfiltered changelog length for transaction start
>>>
>>> Previously we used the normal changelog length for the transaction start.
>>> This
>>> triggered the computehidden logic which in a future patch triggers logic
>>> in the
>>> new hidden store which may start a transaction. This is circular and
>>> results in
>>> broken transaction mechanics, so let's use the unfiltered repo instead to
>>> avoid
>>> triggering computehidden during transaction start.
>>>
>>> This seems like how it should've been in the first place anyway since
>>> recording
>>> the length of the filtered changelog seems odd for a transaction.
>>>
>>> This was causing test failures in a future patch, so test coverage will
>>> come
>>> from the future patch.
>>
>>
>> yes, writing filtered journal seems a bit odd. However I would rather see
>> the whole _writejournal run unfiltered.
>>
>> There are a '@unfilteredmethod' decorator for this purpose.
>
> That seems like a trivial little improvement for V2. There are no
> other calls in the method that seem like they would be affected, so it
> should be functionally equivalent as far as I can tell.

Yep, seems fine as a follow up.

Cheers,

-- 
Pierre-Yves David