Skip to contentSkip to conversation

Is that an acceptable interface change?

DISCUSSION Olivier Galibert 6 participants 9 messages Mar 6, 2006

lore
Share & tools
lore
#0 lore raw

I'm looking at the changes in the asound.h file, and especially at commit 512bbd6a85230f16389f0dd51925472e72fc8a91, and I've been wondering if it's acceptable compatibility-wise. All the structures passed through ioctl (and ALSA is 100% ioctl) have been renamed from sndrv_* to snd_*. That breaks source compatibility but not binary compatibility.

Ignoring the fact that the changelog comment utterly fails to mention that part of the change, is keeping compatibility supposed to be binary-only or source too?

OG.

#1 lore raw
Quoted 6 lines · Olivier Galibert

I'm looking at the changes in the asound.h file, and especially at commit 512bbd6a85230f16389f0dd51925472e72fc8a91, and I've been wondering if it's acceptable compatibility-wise. All the structures passed through ioctl (and ALSA is 100% ioctl) have been renamed from sndrv_* to snd_*. That breaks source compatibility but not binary compatibility.

only if you are "stupid" enough to use kernel headers in your userspace! Which you shouldn't do normally

#2 lore raw
Quoted 10 lines · Arjan van de Ven
Quoted 6 lines · Olivier Galibert

I'm looking at the changes in the asound.h file, and especially at commit 512bbd6a85230f16389f0dd51925472e72fc8a91, and I've been wondering if it's acceptable compatibility-wise. All the structures passed through ioctl (and ALSA is 100% ioctl) have been renamed from sndrv_* to snd_*. That breaks source compatibility but not binary compatibility.

only if you are "stupid" enough to use kernel headers in your userspace! Which you shouldn't do normally

Does that mean it is the responsability of whoever packages the headers for userspace consumption to rename the structs back? Or that every application should come with its own copy of the kernel headers it may need and be ready for massive source-level breakage when rebasing?

I'm just trying to understand if we care about source compatibility for userspace or not.

OG.

#3 lore raw
Quoted 12 lines · Olivier Galibert
Quoted 10 lines · Arjan van de Ven
Quoted 6 lines · Olivier Galibert

I'm looking at the changes in the asound.h file, and especially at commit 512bbd6a85230f16389f0dd51925472e72fc8a91, and I've been wondering if it's acceptable compatibility-wise. All the structures passed through ioctl (and ALSA is 100% ioctl) have been renamed from sndrv_* to snd_*. That breaks source compatibility but not binary compatibility.

only if you are "stupid" enough to use kernel headers in your userspace! Which you shouldn't do normally

Quoted 9 lines

Does that mean it is the responsability of whoever packages the headers for userspace consumption to rename the structs back? Or that every application should come with its own copy of the kernel headers it may need and be ready for massive source-level breakage when rebasing?

I'm just trying to understand if we care about source compatibility for userspace or not.

Userspace apps should not include kernel headers, period. So, userspace applications really shouldn't care.

--
Jesper Juhl <[email protected]>
Don't top-post http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please http://www.expita.com/nomime.html

#4 lore raw
Quoted 2 lines · Jesper Juhl

Userspace apps should not include kernel headers, period. So, userspace applications really shouldn't care.

Please excuse me if I'm a little dense here, but the kernel headers _define_ the userspace-kernel interface. If you don't have them or a sanitized copy of them you just can't talk with the kernel at all.

OG.

#5 lore raw
Quoted 8 lines · Olivier Galibert
Quoted 2 lines · Jesper Juhl

Userspace apps should not include kernel headers, period. So, userspace applications really shouldn't care.

Please excuse me if I'm a little dense here, but the kernel headers _define_ the userspace-kernel interface. If you don't have them or a sanitized copy of them you just can't talk with the kernel at all.

Well, stuff like glibc, the alsa lib etc are exceptions, but the exceptions are few and far between.
Random userspace apps should not use the kernel headers directly, they should instead talk to things like glibc and let glibc handle the userspace<->kernel bit.

--
Jesper Juhl <[email protected]>
Don't top-post http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please http://www.expita.com/nomime.html

#6 lore raw
In reply to Olivier Galibert
Quoted 7 lines · On Mon, 6 Mar 2006 17:15:12 +0100 Olivier Galibert wrote:
Quoted 2 lines · Jesper Juhl

Userspace apps should not include kernel headers, period. So, userspace applications really shouldn't care.

Please excuse me if I'm a little dense here, but the kernel headers _define_ the userspace-kernel interface. If you don't have them or a sanitized copy of them you just can't talk with the kernel at all.

JJ should/could have said what to use instead.*
You should use glibc sanitized kernel headers.

Raw kernel headers are not appropriate or supported for use in userspace.

*: similar to some replies like "No" without further explanation.

---
~Randy

#7 lore raw
In reply to Olivier Galibert
Quoted 2 lines

Please excuse me if I'm a little dense here, but the kernel headers _define_ the userspace-kernel interface.

I don't think so. The kernel headers define the API's that the various kernel facilities present to each other, within the kernel.

The userspace-kernel interface is not really a simple interface of procedures and types definable by a traditional C header. It uses a variety of techniques, such as special calling conventions, special files, and what not ... almost everything possible except the simple call by one procedure of another on a common stack.

In the general case, we can do no more than document, from the kernel side, what is the interface, and expect libraries and such on the user side to wrap this up in a form palatable to C (and other diverse) language uses.

Such header files as are useful to a userspace C programmer are usually provided by the userspace library, for it is that library that the application is linking with, not the kernel.

Signature · 3 lines
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <[email protected]> 1.925.600.0401
#8 lore raw
In reply to Jesper Juhl
Quoted 2 lines · Jesper Juhl

Userspace apps should not include kernel headers, period. So, userspace applications really shouldn't care.

Yeah, but doesn't changing the names like this just make more work for the guys that sanitize the kernel headers and have to stay source-compatible with previous versions?

I mean, we shouldn't make extra work for other projects just for fun...

Chris