Skip to contentSkip to conversation

Simple replacement of an expression of given type?

Cocci Rusty Russell 3 participants 4 messages Jun 20, 2014 – Jun 21, 2014

lore
Share & tools
lore
#0 lore raw

Hi!

In my current project, I moved a field from a structure
(struct block) into another network-endian structure. I used the following patch.cocci to do the fixups:

        @ rule1 @
        struct block *b;
        @@
        
        - b->blocknum
        + le32_to_cpu(b->hdr->depth)

It got 99%, but it missed the two cases where b was actually in another structure, ie. "te->block->blocknum".

I *think* this is because I want to substitute any expression of type struct block *, rather than only identifiers?

But I couldn't understand the documentation enough to do this :(

Clues appreciated!
Rusty.

#1 lore raw
Quoted 13 lines · Rusty Russell

Hi!

In my current project, I moved a field from a structure
(struct block) into another network-endian structure. I used the following patch.cocci to do the fixups:

         @ rule1 @
         struct block *b;
         @@

         - b->blocknum
         + le32_to_cpu(b->hdr->depth)
Quoted 7 lines

It got 99%, but it missed the two cases where b was actually in another structure, ie. "te->block->blocknum".

I *think* this is because I want to substitute any expression of type struct block *, rather than only identifiers?

But I couldn't understand the documentation enough to do this :(

This should work. But by default coccinelle does not parse global headers, so if the struct for te is defined in a global header it won't know that te->block is of type 'struct block *'. If that is the case try to run it with --all-includes

- Lars

#2 lore raw
In reply to Rusty Russell
Quoted 13 lines · Rusty Russell

Hi!

In my current project, I moved a field from a structure
(struct block) into another network-endian structure. I used the following patch.cocci to do the fixups:

        @ rule1 @
        struct block *b;
        @@
        
        - b->blocknum
        + le32_to_cpu(b->hdr->depth)
Quoted 7 lines

It got 99%, but it missed the two cases where b was actually in another structure, ie. "te->block->blocknum".

I *think* this is because I want to substitute any expression of type struct block *, rather than only identifiers?

But I couldn't understand the documentation enough to do this :(

You are already working on expressions. The problem is that sometimes Coccinelle doesn't know what is the type of te->block. To give it the most possible amount of informations, you can give the options

--recursive-includes --relax-include-path

The first means that it will take into account include files that are referenced from other include files rather than just those referenced directly in the C file. The second means that if it doesn't find xyz.h in the include path but it does find exactly one xyz.h somewhere else in the source tree, then it will use that xyz.h, since it has to be the right one.

This will probably take more time than it is taking now, because it has to parse more code. You may want to use the option --use-cache so that it will cache already parsed code.

julia

#3 lore raw
Quoted 21 lines · Julia Lawall
Quoted 19 lines · Rusty Russell

Hi!

In my current project, I moved a field from a structure
(struct block) into another network-endian structure. I used the following patch.cocci to do the fixups:

        @ rule1 @
        struct block *b;
        @@
        
        - b->blocknum
        + le32_to_cpu(b->hdr->depth)

It got 99%, but it missed the two cases where b was actually in another structure, ie. "te->block->blocknum".

I *think* this is because I want to substitute any expression of type struct block *, rather than only identifiers?

Quoted 7 lines
Quoted 1 line

But I couldn't understand the documentation enough to do this :(

You are already working on expressions. The problem is that sometimes Coccinelle doesn't know what is the type of te->block. To give it the most possible amount of informations, you can give the options

--recursive-includes --relax-include-path

Ah, thanks for the explanation, and indeed, that worked perfectly.

And thanks Julia and Lars for your quick responses!

Cheers,
Rusty.