Skip to contentSkip to conversation

Blacklist binary-only modules lying about their license

PATCH Robert M. Stockmann 3 participants 6 messages Apr 27, 2004 – May 1, 2004 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw
Quoted 14 lines · Michael Poole

"Robert M. Stockmann" <[email protected]> writes:

[snip]

Quoted 10 lines

To be more specific about what i am complaining about, here's a error message i get when doing ./configure inside ntfsprogs-1.8.4 :

checking version of gcc... 2.95.3, bad
configure: error: Please upgrade your gcc compiler to gcc-2.96+ or gcc-3+ version! Earlier compiler versions will NOT work as these do not support unnamed/annonymous structures and unions which are used heavily in linux-ntfs. [jackson:stock]:(~/src/ntfsprogs-1.8.4)$

Aha, unnamed/annonymous structures and unions .....

Quoted 7 lines

Please keep rants about applications' coding style where they belong:
off the linux-kernel list. If you bothered to READ what Linus wrote, you would see that the structure being defined has a name -- in fact, gcc requires that, since the structure would be defined at file scope rather than inside another structure or union.

Michael

Hi Michael,

look i have made complaints about gcc-3.x some time ago, on the gcc mailinglist. Also there they put my opinions aside, with arguments like any powerfull feature can be used in a bad and in a good way. The powerfull feature here is the C99 coding style, which allows for unnamed and anonymous structures and unions. Don't kill our C99 cause it can do bad things. Of course not.

However when dealing with OEM hardware vendors, which have signed contracts with Microsoft about windows drivers, i think we cannot allow these same OEM hardware vendors to introduce binary only drivers of unknown quality into the linux kernel. Its exactly these events which give the Linux kernel, or even Linux in general, give a bad reputation. Think about rants like this :

"The Promise FastTrak TX4000 IDE-RAID controller works perfect inside Windows,
but inside Linux i even lost my data."

The Promise Fasttrak issue i discussed here :

http://gcc.gnu.org/ml/gcc/2004-02/msg00293.html

If every major hardware vendor (like e.g. Adaptec, LSI Logic) will change its policy, to implement its drivers as semi- binary only kernel modules, like Promise did with its FastTrak line of controllers, like in the example above, the Open Source lable of the linux kernel can be placed into the computer museum. Isn't that exactly what a certain Redmond software company wants to achieve?

Cheers,

Robert

Signature · 3 lines
Robert M. Stockmann - RHCE
Network Engineer - UNIX/Linux Specialist
crashrecovery.org  [email protected]
#2 lore raw
Quoted 7 lines · Robert M. Stockmann

look i have made complaints about gcc-3.x some time ago, on the gcc mailinglist. Also there they put my opinions aside, with arguments like any powerfull feature can be used in a bad and in a good way. The powerfull feature here is the C99 coding style, which allows for unnamed and anonymous structures and unions. Don't kill our C99 cause it can do bad things. Of course not.

Quoted 6 lines

If every major hardware vendor (like e.g. Adaptec, LSI Logic) will change its policy, to implement its drivers as semi- binary only kernel modules, like Promise did with its FastTrak line of controllers, like in the example above, the Open Source lable of the linux kernel can be placed into the computer museum. Isn't that exactly what a certain Redmond software company wants to achieve?

What the hell do these two paragraphs have to do with each other?

#3 lore raw
Quoted 8 lines · Tim Hockin
Quoted 6 lines · Robert M. Stockmann

look i have made complaints about gcc-3.x some time ago, on the gcc mailinglist. Also there they put my opinions aside, with arguments like any powerfull feature can be used in a bad and in a good way. The powerfull feature here is the C99 coding style, which allows for unnamed and anonymous structures and unions. Don't kill our C99 cause it can do bad things. Of course not.

Quoted 9 lines
Quoted 6 lines

If every major hardware vendor (like e.g. Adaptec, LSI Logic) will change its policy, to implement its drivers as semi- binary only kernel modules, like Promise did with its FastTrak line of controllers, like in the example above, the Open Source lable of the linux kernel can be placed into the computer museum. Isn't that exactly what a certain Redmond software company wants to achieve?

What the hell do these two paragraphs have to do with each other?

C99 coding style, more specific the use of unnamed and anonymous structures and unions, allows the kernel programmer to interface, read glue, binary only driver modules to interface with any linux kernel source tree.

Using this feature one can link and merge any binary only module.o to any kernel source version of your choice. In this way, typically, a OEM vendor, releases a semi open-source link kit, which also contains binary only components, once. This semi open source link-kit will then work with any 2.6.x or 2.4.2x kernel source tree.

The needed header files, which need to be read by the gcc compiler, contain unnamed and annonymizes structures and unions. In the worst case scenario, only the name of used variables are given and no info about variable type or size are inside these headers files. gcc-2.95.3 fails to succesfully link these semi open-source link-kits, and gcc-3.x (which supports C99) of course has no problems doing this.

Robert

Signature · 3 lines
Robert M. Stockmann - RHCE
Network Engineer - UNIX/Linux Specialist
crashrecovery.org  [email protected]
#4 lore raw
Quoted 5 lines · Robert M. Stockmann
Quoted 1 line

What the hell do these two paragraphs have to do with each other?

C99 coding style, more specific the use of unnamed and anonymous structures and unions, allows the kernel programmer to interface, read glue, binary only driver modules to interface with any linux kernel source tree.

What the hell are you going on about? Unnamed structures are a syntactical construct and have ZILCH to do with runtime.

Quoted 4 lines

The needed header files, which need to be read by the gcc compiler, contain unnamed and annonymizes structures and unions. In the worst case scenario, only the name of used variables are given and no info about variable type or size are inside these headers files. gcc-2.95.3 fails to succesfully link these

Opaque types have been available FOREVER.

#5 lore raw
Quoted 9 lines · Tim Hockin
Quoted 5 lines · Robert M. Stockmann
Quoted 1 line

What the hell do these two paragraphs have to do with each other?

C99 coding style, more specific the use of unnamed and anonymous structures and unions, allows the kernel programmer to interface, read glue, binary only driver modules to interface with any linux kernel source tree.

What the hell are you going on about? Unnamed structures are a syntactical construct and have ZILCH to do with runtime.

I thought so too, until your semi open-source link kit is linked to that brand-new linux kernel source tree, and at the same time the binary components of your link-kit have become incompatible with that newer kernel.

Result? one might even loose its data, upon booting that newly build kernel and modules, in case your storage-controller has a binary only link-kit as its driver.

Quoted 6 lines
Quoted 4 lines

The needed header files, which need to be read by the gcc compiler, contain unnamed and annonymizes structures and unions. In the worst case scenario, only the name of used variables are given and no info about variable type or size are inside these headers files. gcc-2.95.3 fails to succesfully link these

Opaque types have been available FOREVER.

sure, but can one qualify that as Open Source?

Robert

Signature · 3 lines
Robert M. Stockmann - RHCE
Network Engineer - UNIX/Linux Specialist
crashrecovery.org  [email protected]
#6 lore raw
Quoted 8 lines · On Apr-28 2004, Wed, 01:59 +0200 Robert M. Stockmann <[email protected]> wrote:
Quoted 6 lines
Quoted 4 lines

The needed header files, which need to be read by the gcc compiler, contain unnamed and annonymizes structures and unions. In the worst case scenario, only the name of used variables are given and no info about variable type or size are inside these headers files. gcc-2.95.3 fails to succesfully link these

Opaque types have been available FOREVER.

sure, but can one qualify that as Open Source?

Are you familiar with my friend "void *"?

Signature · 1 line
Tomas Szepe <[email protected]>