Re: Discriminants or Constructor Function for Limited Types

["Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de>]

On 2022-05-09 14:05, Doctor Who wrote:
> On Mon, 9 May 2022 13:15:54 +0200, "Dmitry A. Kazakov"
> <mailbox@dmitry-kazakov.de> wrote:
> 
>> On 2022-05-09 12:19, Doctor Who wrote:
>>
>>> if you do a check before write there is no need to rise and propagate
>>> an exception.
>>
>> A bad idea.
>>
>> First it is an unnecessary overhead, because ultimately the check will
>> be repeated.
>>
>> Secondly it is technically impossible to do for a huge number of reasons:
>>
>> 1. Too complex to do. The modern hardware and software does a lot of
>> bookkeeping, indexing, replicating, compression, encryption, the stuff
>> almost impossible to estimate or predict in advance.
>>
>> 2. Unreliable due to racing conditions, volatile network states etc.
>>
>> 3. Very expensive, e.g. in the case of a networking file system or a
>> database and the asynchronous nature of modern I/O subsystems. When you
>> write a file you normally do not wait for the operation completion.
>> Actual writing continues asynchronously to your application going
>> through dozens of stacks, caches, buffers, buses. If that fails you will
>> learn that later in a consequent operation unless you explicitly call
>> Flush or equivalent. Checking the state is a synchronous operation that
>> would have hundred- to thousandfold performance penalty.
>>
>> 4. Just impossible like when dealing with a stream, a pipe line etc.
>>
>> As general design rules regarding exceptions consider:
>>
>> 1. Never use return code.
>>
>> 2. Never check anything non-trivial in advance.
> 
> we are talking about files, in a local disk right?

Yes, all this fully applies to local disks and filesystems.

Just look how writing an SSD works. I think you could get better chances 
with magnetic drums from 60's. (:-))

-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de