Subject: Re: brokenness
To: Noriyuki Soda <soda@sra.co.jp>
From: Castor Fu <castor@geocast.net>
List: port-mips
Date: 01/19/1999 07:30:45
I asked about the idea of <machine/pubassym.h> on three separate
occassions over the course of more than a month.  I received absolutely
zero comments on this.  I described this once on developers, once on
port-mips and once by providing the code for review for a week prior to
committing it.  While this certainly does not mean that one forfeits
the right to comment on changes, a a certain civility is appropriate when
commenting at such a late date.

Dealing with attitudes like this is what causes people and
corporations to invest their time and support elsewhere instead of
in NetBSD.

It would be helpful to point out why something breaks a standard to
help create a solution.  I presume the POSIX/ANSI C issues  concern
namespace pollution.  Is using a single underscore prefix sufficient?
Double underscores?  Trailing underscores?

That said, I'll probably be moving the checking code to regress where 
any discrepancies will cause a build break, and machine/pubassym.h will
have roughly the same set of symbols except prefaced by underscores,
or maybe by _JB.

Constructive criticism is welcome.

	-castor

On Tue, 19 Jan 1999, Noriyuki Soda wrote:

> Date: Tue, 19 Jan 1999 19:42:36 +0900 (JST)
> From: Noriyuki Soda <soda@sra.co.jp>
> To: castor@netbsd.org, nisimura@netbsd.org
> Cc: port-mips@netbsd.org
> Subject: <machine/pubassym.h> brokenness
> 
> First of all, thanks Mr. castor and Nisimura-san for recent great
> changes for sys/arch/mips.
> 
> But that seems to introduce some minor brokenness to the source tree.
> 
> I think the <machine/pubassym.h> is NOT correct way to deal with _JBLEN.
> Please back out the pubassym.h related changes, and remove the bogons
> about pubassym.h.
> 
> Because,
> 
> - There are no need to automatically calculate _JBLEN.
>   If _JBLEN is changed, then,
> 	libc major version should be changed
> 	(or, at least, libc symbols should be renamed by __RENAME())
>   So, the changes about _JBLEN should be MANUALLY handled.
>   It is good to introduce the utility to check the sanity of _JBLEN,
>   but there is no need to automatically calculate it.
> 
>   So, PLEASE DO NOT AUTOMATICALLY CALCULATE IT !!!
>   IT BREAKS BINARY COMPATIBILTY !!!
> 
> - <machine/pubassym.h> introduce some symbols like STAND_FRAME_SIZE,
>   STAND_RA_OFFSET, CALLFRAME_SIZ, CALLFRAME_SP, CALLFRAME_RA to 
>   <setjmp.h>, So, apparently this breaks the ANSI/ISO-C standard
>   specification and POSIX standard specification. <setjmp.h> should
>   NOT define non-standard symbols like these.
> 
> Please remove <machine/pubassym.h>, thanks.