uima-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Marshall Schor <...@schor.com>
Subject Re: Get all reachable FSes / all FSes?
Date Fri, 11 Sep 2015 21:58:33 GMT
+1.  There's code that does just what Thilo says, used in binary serialization,
I think.  -Marshall

On 9/10/2015 9:15 AM, Thilo Goetz wrote:
> On 09/10/2015 09:17 AM, Richard Eckart de Castilho wrote:
>> On 10.09.2015, at 08:46, Richard Eckart de Castilho <rec@apache.org> wrote:
>>
>>> On 10.09.2015, at 04:06, Marshall Schor <msa@schor.com> wrote:
>>>
>>>> I don't think there's support for the 2nd, directly.
>>>>
>>>> It's doable.  There are various bits of code around that sequentially walk
the
>>>> heap, which is what this would amount to, I think.
>>>>
>>>> I'm curious about the need for such a capability?
>>>
>>> Basically for doing sanity checks on the CAS, e.g. to detect if client code
>>> creates FSes but never adds them to any index.
>>>
>>> We've hit a bug in one piece of software where some annotations were indexed
>>> an others not. The latter were still reachable in that case so I could
>>> implement the sanity check without this capability. But while doing so, I
>>> was wondering whether it was also possible to recover non-indexed FSes.
>>
>> Another case would be detecting FSes that become unreachable unintentionally.
>>
>> Cheers,
>>
>> -- Richard
>>
>
> This used to be easy when I still worked on the code, and it probably still
> is. The heap starts at position 1 (position 0 is the reserved null reference).
> The first positon of a heap entry is the type of the FS. There is a table that
> tells you how much space an entry of a certain type takes. So you know where
> the next FS starts. And so on. As soon as you reach a type cell that contains
> a 0, you know you have reached the end of the occupied heap cells. That way
> you can look at all FSs that were ever written to the heap.
>
> There are low-level APIs that let you instantiate FSs from a given heap
> position (but you probably knew that).
>
> --Thilo
>
>


Mime
View raw message