Re: eCos 3.0 beta 1 punch list #2

[Jonathan Larmour <jifl at eCosCentric dot com>]

Bart Veer wrote:
> In theory there is one more change to go in: updating the flash
> subsystem to use a prioritized constructor, and associated API
> changes. Effectively this would replace explicit calls of
> cyg_flash_init() with automatic initialization via a C++ static
> constructor. That should simplify higher-level code since there is no
> longer any need to worry about whether or not the flash subsystem has
> been initialized. It would also give some code size savings in the
> flash subsystem, and in the medium to long term it would allow other
> subsystems such as fconfig to be statically initialized as well.
>
> However, after experimenting with various different patches, I have
> come to the conclusion that this is not the right time to make the
> change. It is not too bad when there is only a single flash device and
> everything works, but if there are initialization problems then things
> get messy.

I need more detail than "get messy". This is what the dev->init member is 
meant to be for. Uninitialised devices get subsequently ignored, and the 
return value of cyg_flash_init is irrelevant.

As with the spec for CYGBLD_GLOBAL_WARNINGS, you seem to be unilaterally 
moving the goalposts from what had previously been agreed *extremely* late 
in the day.

> So instead of making the change now, for all anoncvs
> targets and with no possiblity of testing on most of those targets, I
> want to do the work in eCosPro first. It can then be merged into a
> future anoncvs release, once I am confident that it is not going to
> break anything.

Why can it not be in both? Why should eCosPro be special with respect to a 
changing API? And incompatible.

I don't know whether to wear an eCosCentric hat which says that eCosPro 
should be aimed at well tested code, rather that a testbed for allegedly 
unstable new code; or wear a maintainer hat and say that what happens in 
eCosPro is not relevant to the public project, and API incompatibilities 
made in such a way as to place eCosPro as "more advanced" is not a 
principle to be encouraged.

> 2) the specification of cyg_flash_init(), i.e. whether or not it
>    takes a printf() function as argument, is not actually important.
>    cyg_flash_init() will become deprecated, and mostly a no-op, when
>    the flash subsystem switches to using a prioritized constructor.
>    It does not actually matter if a deprecated function takes a
>    function pointer as argument.

But you've completely omitted how to properly set the printf function.

>    That means there are no long-term API concerns either. There are
>    functions to be added to the API, but those can wait till later.

I disagree. This means we are issuing an API which we have to instantly 
deprecate and anyone presently coding to that API will encounter breakage 
later.

If the API is wrong we need to fix out before we make a major release. A 
stable flash API was the second most critical change to make for eCos 3.0 
after the header updates. We can't flub it. This is a problem I raised 
with you back in 2006.

Let me know what you consider the problems to be and I will deal with them.

Jifl
--
*See us at Embedded World 2009, Nürnberg, Germany, 3-5 Mar, Stand 11-300*
eCosCentric Limited      http://www.eCosCentric.com/     The eCos experts
Barnwell House, Barnwell Drive, Cambridge, UK.       Tel: +44 1223 245571
Registered in England and Wales: Reg No 4422071.
------["Si fractum non sit, noli id reficere"]------       Opinions==mine