bpf_dsl: support Switch/Case expressions

bpf_dsl: support Switch/Case expressions

This adds support for expressions like

    Switch(arg)
      .Case(1, result1)
      .CASES((2, 3), result2)
      .Default(result3)

Currently these expressions are compiled simply as short-hand for

    If(arg == 1, result1)
      .ElseIf(arg == 2 || arg == 3, result2)
      .Else(result3)

but in the future we should be able to optimize it better.

The CASES macro ugliness is unfortunately necessary until we're
allowed to use C++11.  Later we'll be able to change it to
real C++ code like "Cases({2, 3}, result2)" (which clang-format
will then format a bit more nicely too).

BUG=408845
R=jln@chromium.org, jorgelo@chromium.org

Committed: https://chromium.googlesource.com/chromium/src/+/279b9aad9824c840e10382c2f6e4356d4c632774

[mdempsky, 2014-08-01 09:20:35]

Just sharing as a proof-of-concept.  I like the idea, and the resulting policy
code seems a bit easier to read, but the implementation is a bit hacky and I'm
not sure it's outweighed by the minor gain in conciseness.

Ideally I'd use C++11 initializer lists and Cases() would be called like

    Cases({3, 5, 7}, result)

but to be [GNU] C++03 compatible, for now I'm using some macro tricks and
instead writing the above as

    Cases((3, 5, 7), result)

(Details: The parentheses force the preprocessor to skip over the inner commas
so it sees the two macro arguments as "(3, 5, 7)" and "result"; then when the
Cases macro expands, "CasesHelper values" will become "CasesHelper (3, 5, 7)"
which will trigger another macro expansion while also stripping away the
unwanted parentheses.)

[Jorge Lucangeli Obes, 2014-08-02 00:55:41]

On 2014/08/01 09:20:35, mdempsky wrote:
> Just sharing as a proof-of-concept.  I like the idea, and the resulting policy
> code seems a bit easier to read, but the implementation is a bit hacky and I'm
> not sure it's outweighed by the minor gain in conciseness.
> 
> Ideally I'd use C++11 initializer lists and Cases() would be called like
> 
>     Cases({3, 5, 7}, result)
> 
> but to be [GNU] C++03 compatible, for now I'm using some macro tricks and
> instead writing the above as
> 
>     Cases((3, 5, 7), result)
> 
> (Details: The parentheses force the preprocessor to skip over the inner commas
> so it sees the two macro arguments as "(3, 5, 7)" and "result"; then when the
> Cases macro expands, "CasesHelper values" will become "CasesHelper (3, 5, 7)"
> which will trigger another macro expansion while also stripping away the
> unwanted parentheses.)

IMHO the policies look significantly clear and the extra code is not super ugly.

[mdempsky, 2014-09-03 01:11:20]

I've cleaned this up some, so it should be ready for review and submission. PTAL

[Jorge Lucangeli Obes, 2014-09-03 15:35:19]

lgtm. Arg filters look so much better.

[mdempsky, 2014-09-03 21:36:51]

On 2014/09/03 15:35:19, Jorge Lucangeli Obes wrote:
> lgtm. Arg filters look so much better.

Thanks. :)

jln: Interested in any feedback you might have too, when you get a chance.

[jln (very slow on Chromium), 2014-09-05 22:48:39]

lgtm

I think the compiler optimization for these might be worth it in a few cases
where we do indeed have quite a long list.

https://chromiumcodereview.appspot.com/438683004/diff/100001/sandbox/linux/bp...
File sandbox/linux/bpf_dsl/bpf_dsl.h (right):

https://chromiumcodereview.appspot.com/438683004/diff/100001/sandbox/linux/bp...
sandbox/linux/bpf_dsl/bpf_dsl.h:344: for (Iter i = values.begin(), end =
values.end(); i != end; i++) {
Nit: I find ++i more idiomatic (because there is no need to keep a temporary).
It doesn't matter, so your choice.

[mdempsky, 2014-09-05 23:08:31]

https://chromiumcodereview.appspot.com/438683004/diff/100001/sandbox/linux/bp...
File sandbox/linux/bpf_dsl/bpf_dsl.h (right):

https://chromiumcodereview.appspot.com/438683004/diff/100001/sandbox/linux/bp...
sandbox/linux/bpf_dsl/bpf_dsl.h:344: for (Iter i = values.begin(), end =
values.end(); i != end; i++) {
On 2014/09/05 22:48:39, jln wrote:
> Nit: I find ++i more idiomatic (because there is no need to keep a temporary).
> It doesn't matter, so your choice.

Done.  (Oops, yeah, I usually use preincrement too.)

[mdempsky, 2014-09-06 01:19:33]

Committed patchset #7 (id:120001) manually as 279b9aa (presubmit successful).

[commit-bot: I haz the power, 2014-09-10 03:42:07]

Patchset 7 (id:??) landed as
https://crrev.com/279b9aad9824c840e10382c2f6e4356d4c632774
Cr-Commit-Position: refs/heads/master@{#293599}