Quote from overview:

Features To Drop
-Bit fields of arbitrary size. Bit fields are a complex, inefficient feature
rarely used.

 You mention that you want to have 'D' be able to directly interface with
hardware, "...retains the ability to write high performance code and
interface directly with the operating system APIs and with hardware."

Many types of hardware (and file-formats) make use of bit-fields in their structures.  While it is always possible to make masks and bit shift stuff yourself, it is more prone to error than using bit-fields.  I think that bit-fields are a part of the compiler which many people will implement themselves if it's not included (kinda like string class in C++).

-Rolf Campbell

In article <9lgqhk$2clv$1@digitaldaemon.com>, Rolf Campbell wrote:
> Quote from overview:
> 
> Features To Drop
> -Bit fields of arbitrary size. Bit fields are a complex, inefficient feature
> rarely used.
> 
>  You mention that you want to have 'D' be able to directly interface with
> hardware, "...retains the ability to write high performance code and
> interface directly with the operating system APIs and with hardware."

This is one problem.  However, if you have bitfields, please define the ordering that they will have bits in.  Bitfields are not portable in C/C++ due to this fact.  All OS software needs to use masks/etc to write "portable" drivers in C.


> Many types of hardware (and file-formats) make use of bit-fields in their structures.  While it is always possible to make masks and bit shift stuff yourself, it is more prone to error than using bit-fields.  I think that bit-fields are a part of the compiler which many people will implement themselves if it's not included (kinda like string class in C++).

This again is very true.  I would recommend having some sort of bitfield mechanism, along with some mechanism to access memory and I/O ports.  This could well be implemented in modules (using assembly language) on each port of the language.  With automatic inlining (and link-optimization passes), this could be as good, or better than language support for the various "machine oriented" constructs.

In other words, have a bitfield module, that implements both LE, BE, and other ordering for bitfields.


-- 
Tobias Weingartner |        Unix Guru, Admin, Systems-Dude
Apt B 7707-110 St. |        http://www.tepid.org/~weingart/
Edmonton, AB       |-------------------------------------------------
Canada, T6G 1G3    | %SYSTEM-F-ANARCHISM, The OS has been overthrown

Rolf Campbell wrote:

> Quote from overview:
> 
> Features To Drop
> -Bit fields of arbitrary size. Bit fields are a complex, inefficient feature
> rarely used.
> 
>  You mention that you want to have 'D' be able to directly interface with
> hardware, "...retains the ability to write high performance code and
> interface directly with the operating system APIs and with hardware."
> 
> Many types of hardware (and file-formats) make use of bit-fields in their
> structures.  While it is always possible to make masks and bit shift stuff
> yourself, it is more prone to error than using bit-fields.  I think that
> bit-fields are a part of the compiler which many people will implement
> themselves if it's not included (kinda like string class in C++).
> 
> -Rolf Campbell
> 
> 
> 

Hear, hear!  I write embedded systems for a living, and most of the configuration
and status registers are implemented with a bit here and a couple of bits there.

This is an extremely useful language feature in doing embedded systems.

Matt

"Matt Gessner" <mattg@aiinet.com> wrote in message news:3B7C19A7.3040203@aiinet.com...
> Rolf Campbell wrote:
> > Quote from overview:
> > Features To Drop
> > -Bit fields of arbitrary size. Bit fields are a complex, inefficient
feature
> > rarely used.
> >
> >  You mention that you want to have 'D' be able to directly interface
with
> > hardware, "...retains the ability to write high performance code and interface directly with the operating system APIs and with hardware."
> >
> > Many types of hardware (and file-formats) make use of bit-fields in
their
> > structures.  While it is always possible to make masks and bit shift
stuff
> > yourself, it is more prone to error than using bit-fields.  I think that bit-fields are a part of the compiler which many people will implement themselves if it's not included (kinda like string class in C++).
> >
> > -Rolf Campbell
> Hear, hear!  I write embedded systems for a living, and most of the
configuration
> and status registers are implemented with a bit here and a couple of bits
there.
>
> This is an extremely useful language feature in doing embedded systems. Matt

When I've done hardware I/O, I always wound up doing masking and shifting, because it produced better code.

Walter wrote:

> When I've done hardware I/O, I always wound up doing masking and shifting, because it produced better code.

Yes, but debugging massive bit twiddling for things like long bit strings to SPI devices quickly becomes a nightmare.  It would truly be beneficial to have the D "bit" type enhanced for packing into fundamental integer types (and arrays of same).

One big problem is that the same D code must compile and run on platforms with different endian-ness!  The same hardware chip (say, an Ethernet controller) with a mix of 8/16/32-bit registers soon becomes a nightmare, depending on if the device is accessed by byte, word or dword.  The situation in C often requires special macros to handle endian-ness, or (worse) a mess of conditionally compiled code.

Another embedded perspective:  It would be good to define ways of writing D code that will GUARANTEE the garbage collector will NOT be invoked!  Perhaps a compiler switch to disable both GC and the language features that rely upon it.  I'd willing accept a restricted language subset for certain circumstances (say, device drivers and interrupt handlers) if it will allow me to program an entire system in a single language.

When using C, I often result to inline assembly ("asm(...)") to handle cases
the language can't handle well.  But this makes the code inherently
non-portable, so I always wind up including the messy C equivalent in a
#ifdef'ed code section ("#ifdef NOASM ...").


-BobC