qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH 0/8] OptsVisitor: support / flatten integer rang


From: Laszlo Ersek
Subject: Re: [Qemu-devel] [PATCH 0/8] OptsVisitor: support / flatten integer ranges for repeating options
Date: Fri, 16 Aug 2013 09:15:51 +0200
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:17.0) Gecko/20130806 Thunderbird/17.0.8

Ping then, for 1.7 :)

(Don't stone me, I don't have the slightest clue about the 1.7 backlog.
I'm fine with this being tacked to the end, just don't let it fall
through the cracks.)

Thanks
Laszlo

On 07/22/13 23:07, Laszlo Ersek wrote:
> rfc->v1:
> - addressed Paolo's comments for patches 1 and 2,
> - patches 7 and 8 are new (unit tests),
> - updated the cover letter to take native lists into account, plus
>   cleaned it up.
> 
> Consider the following QAPI schema fragment, for the purpose of command
> line parsing with OptsVisitor:
> 
>   { 'union': 'NumaOptions',
>     'data': {
>       'node': 'NumaNodeOptions',
>       'mem' : 'NumaMemOptions' }}
> 
>   { 'type': 'NumaNodeOptions',
>     'data': {
>       '*nodeid': 'int',
>       '*cpus'  : ['uint16'] }}
> 
>   { 'type': 'NumaMemOptions',
>     'data': {
>       '*nodeid': 'int',
>       '*size'  : 'size' }}
> 
> (Commit eb7ee2cb ("qapi: introduce OptsVisitor") had originally
> documented OptsVisitor's general schema requirements for parsing
> repeated options such that the list element type had to be a struct with
> one mandatory scalar field. Accordingly, the RFC version of this series
> required for interval flattening that this underlying scalar type be an
> integer type. However, since commit a678e26c ("qapi: pad GenericList
> value fields to 64 bits") we've had reliable native lists; OptsVisitor
> turns out to support them automatically.)
> 
> OptsVisitor already accepts the following command line with the above
> schema:
> 
>   -numa node,nodeid=3,cpus=0,cpus=1,cpus=2,cpus=6,cpus=7,cpus=8
> 
> Paolo suggested in
> <http://thread.gmane.org/gmane.comp.emulators.qemu/222589/focus=222732>
> that OptsVisitor should allow the following shortcut:
> 
>   -numa node,nodeid=3,cpus=0-2,cpus=6-8
> 
> and that the code processing the "cpus" list should encounter all six
> elements (0, 1, 2, 6, 7, 8) individually.
> 
> The series implements this feature. Both signed and unsigned values and
> intervals are supported in general:
> 
>   * 0     (zero)
>   * 1-5   (one to five)
>   * 4-4   (four to four, range with one element)
>   * -2    (minus two)
>   * -5-8  (minus five to plus eight)
>   * -9--6 (minus nine to minus six)
> 
> The restrictions imposed by the native list element's signedness and
> size (in the above schema example, 'uint16') are enforced element-wise
> as usual. That is, for 'uint16', the command line option
> 
>   -numa node,nodeid=3,cpus=65534-65537
> 
> is equivalent to
> 
>   -numa node,nodeid=3,cpus=65534,cpus=65535,cpus=65536,cpus=65537
> 
> and visit_type_uint16() [qapi/qapi-visit-core.c] will catch the first
> element (= 65536) that has been parsed by opts_type_int() but cannot be
> represented as 'uint16'.
> 
> Laszlo Ersek (8):
>   OptsVisitor: introduce basic list modes
>   OptsVisitor: introduce list modes for interval flattening
>   OptsVisitor: opts_type_int(): recognize intervals when LM_IN_PROGRESS
>   OptsVisitor: rebase opts_type_uint64() to parse_uint_full()
>   OptsVisitor: opts_type_uint64(): recognize intervals when
>     LM_IN_PROGRESS
>   OptsVisitor: don't try to flatten overlong integer ranges
>   add "test-int128" to .gitignore
>   OptsVisitor: introduce unit tests, with test cases for range
>     flattening
> 
>  tests/Makefile              |    6 +-
>  qapi-schema-test.json       |   15 +++
>  include/qapi/opts-visitor.h |    6 +
>  qapi/opts-visitor.c         |  184 ++++++++++++++++++++++++-----
>  tests/test-opts-visitor.c   |  275 
> +++++++++++++++++++++++++++++++++++++++++++
>  .gitignore                  |    2 +
>  6 files changed, 456 insertions(+), 32 deletions(-)
>  create mode 100644 tests/test-opts-visitor.c
> 
> 




reply via email to

[Prev in Thread] Current Thread [Next in Thread]