Can we sort out the prototypes within the cifs headers?
David Howells
maintainerHi Paulo, Enzo, et al.,
You may have seen my patch:
https://lore.kernel.org/linux-cifs/[email protected]/T/#u
to sort out the cifs header file prototypes, which are a bit of a mess: some seem to have been placed haphazardly in the headers, some have unnamed arguments and also sometimes the names in the .h and the .c don't match.
Now Steve specifically namechecked you two as this will affect the backporting of patches. Whilst this only affects the prototypes in the headers and not the implementations in C files, it does cause chunks of the headers to move around.
Can we agree on at least a subset of the cleanups to be made? In order of increasing conflictiveness, I have:
(1) Remove 'extern'. cifs has a mix of externed and non-externed, but the documented approach is to get rid of externs on prototypes.
(2) (Re)name the arguments in the prototypes to be the same as in the implementations.
(3) Adjust the layout of each prototype to match the implementation, just with a semicolon on the end. My script partially does this, but moves the return type onto the same line as the function name.
(4) Move SMB1-specific functions out to smb1proto.h. Move SMB2/3-specific functions out to smb2proto.h.
(5) Divide the lists of prototypes (particularly the massive one in cifsproto.h) up into blocks according to which .c file contains the implementation and preface each block with a comment that indicates the name of the relevant .c file.
The comment could then be used as a key for the script to maintain the division in future.
(6) Sort each block by position in the .c file to make it easier to maintain them.
A hybrid approach is also possible, where we run the script to do the basic sorting and then manually correct the output.
David
Quoted 1 line
(4) Move SMB1-specific functions out to
smb1proto.h. Move SMB2/3-specific
functions out to smb2proto.h.
I am generally in favor of those type of cleanup patches (as potentially higher priority) as we want to be able to turn off/remove SMB1 code easily and not confuse old, less secure SMB1, with modern dialects
Quoted 49 lines · David Howells
Hi Paulo, Enzo, et al.,
You may have seen my patch:
https://lore.kernel.org/linux-cifs/[email protected]/T/#uto sort out the cifs header file prototypes, which are a bit of a mess: some seem to have been placed haphazardly in the headers, some have unnamed arguments and also sometimes the names in the .h and the .c don't match.
Now Steve specifically namechecked you two as this will affect the backporting of patches. Whilst this only affects the prototypes in the headers and not the implementations in C files, it does cause chunks of the headers to move around.
Can we agree on at least a subset of the cleanups to be made? In order of increasing conflictiveness, I have:
(1) Remove 'extern'. cifs has a mix of externed and non-externed, but the documented approach is to get rid of externs on prototypes.
(2) (Re)name the arguments in the prototypes to be the same as in the implementations.
(3) Adjust the layout of each prototype to match the implementation, just with a semicolon on the end. My script partially does this, but moves the return type onto the same line as the function name.
(4) Move SMB1-specific functions out to
smb1proto.h. Move SMB2/3-specific functions out tosmb2proto.h.(5) Divide the lists of prototypes (particularly the massive one in
cifsproto.h) up into blocks according to which .c file contains the implementation and preface each block with a comment that indicates the name of the relevant .c file.The comment could then be used as a key for the script to maintain the division in future.
(6) Sort each block by position in the .c file to make it easier to maintain them.
A hybrid approach is also possible, where we run the script to do the basic sorting and then manually correct the output.
David
Signature · 3 lines
Thanks, Steve
Hi David,
Quoted 36 lines · David Howells
Hi Paulo, Enzo, et al.,
You may have seen my patch:
https://lore.kernel.org/linux-cifs/[email protected]/T/#uto sort out the cifs header file prototypes, which are a bit of a mess: some seem to have been placed haphazardly in the headers, some have unnamed arguments and also sometimes the names in the .h and the .c don't match.
Now Steve specifically namechecked you two as this will affect the backporting of patches. Whilst this only affects the prototypes in the headers and not the implementations in C files, it does cause chunks of the headers to move around.
Can we agree on at least a subset of the cleanups to be made? In order of increasing conflictiveness, I have:
(1) Remove 'extern'. cifs has a mix of externed and non-externed, but the documented approach is to get rid of externs on prototypes.
(2) (Re)name the arguments in the prototypes to be the same as in the implementations.
(3) Adjust the layout of each prototype to match the implementation, just with a semicolon on the end. My script partially does this, but moves the return type onto the same line as the function name.
(4) Move SMB1-specific functions out to
smb1proto.h. Move SMB2/3-specific functions out tosmb2proto.h.(5) Divide the lists of prototypes (particularly the massive one in
cifsproto.h) up into blocks according to which .c file contains the implementation and preface each block with a comment that indicates the name of the relevant .c file.
Quoted 8 lines
The comment could then be used as a key for the script to maintain the division in future.
(6) Sort each block by position in the .c file to make it easier to maintain them.
A hybrid approach is also possible, where we run the script to do the basic sorting and then manually correct the output.
+1 for the cleanups, thanks for doing that.
On backports, I think points 1-3 could be done together, but in separate commits (per header file) to minimise conflicts.
4 looks good to have.
5-6 would be most problematic (moving code around).
Not sure what else to say here, but more atomic commit are easier to backport than big/monolithic ones.
Cheers,
Enzo
I don't have much objection about #1 and $2, 3 possibly if not huge, but higher priority is 4. Agree that 5 and 6 are lowest priority unless part of patch that is fixing (or perf improvement etc.) something
Quoted 66 lines · Enzo Matsumiya
Hi David,
Quoted 36 lines · David Howells
Hi Paulo, Enzo, et al.,
You may have seen my patch:
https://lore.kernel.org/linux-cifs/[email protected]/T/#uto sort out the cifs header file prototypes, which are a bit of a mess: some seem to have been placed haphazardly in the headers, some have unnamed arguments and also sometimes the names in the .h and the .c don't match.
Now Steve specifically namechecked you two as this will affect the backporting of patches. Whilst this only affects the prototypes in the headers and not the implementations in C files, it does cause chunks of the headers to move around.
Can we agree on at least a subset of the cleanups to be made? In order of increasing conflictiveness, I have:
(1) Remove 'extern'. cifs has a mix of externed and non-externed, but the documented approach is to get rid of externs on prototypes.
(2) (Re)name the arguments in the prototypes to be the same as in the implementations.
(3) Adjust the layout of each prototype to match the implementation, just with a semicolon on the end. My script partially does this, but moves the return type onto the same line as the function name.
(4) Move SMB1-specific functions out to
smb1proto.h. Move SMB2/3-specific functions out tosmb2proto.h.(5) Divide the lists of prototypes (particularly the massive one in
cifsproto.h) up into blocks according to which .c file contains the implementation and preface each block with a comment that indicates the name of the relevant .c file.Quoted 8 lines
The comment could then be used as a key for the script to maintain the division in future.
(6) Sort each block by position in the .c file to make it easier to maintain them.
A hybrid approach is also possible, where we run the script to do the basic sorting and then manually correct the output.
+1 for the cleanups, thanks for doing that.
On backports, I think points 1-3 could be done together, but in separate commits (per header file) to minimise conflicts.
4 looks good to have.
5-6 would be most problematic (moving code around).
Not sure what else to say here, but more atomic commit are easier to backport than big/monolithic ones.
Cheers,
Enzo
Signature · 3 lines
Thanks, Steve
David Howells
maintainerQuoted 6 lines · Enzo Matsumiya
+1 for the cleanups, thanks for doing that.
On backports, I think points 1-3 could be done together, but in separate commits (per header file) to minimise conflicts.
4 looks good to have.
Hi Enzo,
Could you have a look at:
https://lore.kernel.org/linux-cifs/[email protected]/T/#t
see if you're okay with it?
Thanks,
David