[Mono-dev] Re: [Fwd: [Mono-patches] r48788 - trunk/mcs/tools/compiler-tester]

[Marek Safar]

Hello,

>> 2005-06-24  Marek Safar  <marek.safar at seznam.cz>
>>
>>>>
>>>>     * compiler-tester.cs: Enabled error/warning message checking.
>>>
>>>
>>>
>>> I think we should never do such kind of check. It will mess
>>> localization of compiler messages in the future. Or do we never
>>> provide localized messages?
>>
>>
>> I think it is important part of regression tests. I had this change 
>> on my box for while and it regularly found some regression errors.
>> I added also missing location check, unfortunately your location 
>> patch was so untested that I had to disable this detection until we 
>> fix mcs.
>
>
> Cool. That feature will expose such incorrect changes that erased
> Location information by someone e.g. incorrect Const replacement.
> Everyone can know what is the cause of the location bug.

Yes, but you will have to uncomment one piece of code in tester code to 
see all regressions.

>
>> I don't see any problem with localization until we will support 
>> English at least one of the error text languages.
>
>
> Until we support English - that means someday this test became
> nothing and then you can't depend on this kind of tests.

I mean we will support English in terms of English will be one of 
localized languages. In such case I still can test English error text 
against expected error texts.

> Similar situations happened on early days of System.Data tests.
> When Microsoft released .NET FX 1.1, both the type of expected
> exceptions and error messages got changed and they highly messed
> System.Data unit tests. I believe this will become messy version
> by version.

I don't think it is same case as we don't copy Microsoft error messages. 
So they can change their texts as they want and we too.

Marek