Re: [PATCH 3/7] oxenstored: refactor request processing

[Jonathan Davies <jonathan.davies@citrix.com>]

On Tue, Mar 29, 2016 at 08:41:54PM +0100, David Scott wrote:
> 
> > On 29 Mar 2016, at 17:38, Wei Liu <wei.liu2@citrix.com> wrote:
> > 
> > On Tue, Mar 29, 2016 at 10:08:30AM +0100, Jonathan Davies wrote:
> >> On Thu, Mar 24, 2016 at 07:57:30PM -0400, Boris Ostrovsky wrote:
> >>> On 03/24/2016 06:49 PM, Andrew Cooper wrote:
> >>>> On 24/03/2016 22:22, Boris Ostrovsky wrote:
> >>>>> On 03/17/2016 01:51 PM, Jonathan Davies wrote:
> >>>>>> Encapsulate the request in a record that is passed from do_input to
> >>>>>> process_packet and input_handle_error.
> >>>>>> 
> >>>>>> This will be helpful when keeping track of the requests made as part
> >>>>>> of a
> >>>>>> transaction.
> >>>>>> 
> >>>>>> Signed-off-by: Jonathan Davies <jonathan.davies@citrix.com>
> >>>>>> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
> >>>>>> Reviewed-by: Jon Ludlam <jonathan.ludlam@citrix.com>
> >>>>>> Reviewed-by: Euan Harris <euan.harris@citrix.com>
> >>>>>> ---
> >>>>>>  tools/ocaml/xenstored/process.ml |   15 ++++++++++-----
> >>>>>>  1 file changed, 10 insertions(+), 5 deletions(-)
> >>>>>> 
> >>>>>> diff --git a/tools/ocaml/xenstored/process.ml
> >>>>>> b/tools/ocaml/xenstored/process.ml
> >>>>>> index 7a73669..c92bec7 100644
> >>>>>> --- a/tools/ocaml/xenstored/process.ml
> >>>>>> +++ b/tools/ocaml/xenstored/process.ml
> >>>>>> @@ -344,11 +344,11 @@ let function_of_type ty =
> >>>>>>      | Xenbus.Xb.Op.Invalid           -> reply_ack do_error
> >>>>>>      | _                              -> reply_ack do_error
> >>>>>>  -let input_handle_error ~cons ~doms ~fct ~ty ~con ~t ~rid ~data =
> >>>>>> +let input_handle_error ~cons ~doms ~fct ~con ~t ~req =
> >>>>>>      let reply_error e =
> >>>>>>          Packet.Error e in
> >>>>>>      try
> >>>>>> -        fct con t doms cons data
> >>>>>> +        fct con t doms cons req.Packet.data
> >>>>>>      with
> >>>>>>      | Define.Invalid_path          -> reply_error "EINVAL"
> >>>>>>      | Define.Already_exist         -> reply_error "EEXIST"
> >>>>>> @@ -370,7 +370,10 @@ let input_handle_error ~cons ~doms ~fct ~ty ~con
> >>>>>> ~t ~rid ~data =
> >>>>>>  (**
> >>>>>>   * Nothrow guarantee.
> >>>>>>   *)
> >>>>>> -let process_packet ~store ~cons ~doms ~con ~tid ~rid ~ty ~data =
> >>>>>> +let process_packet ~store ~cons ~doms ~con ~req =
> >>>>>> +    let ty = req.Packet.ty in
> >>>>>> +    let tid = req.Packet.tid in
> >>>>>> +    let rid = req.Packet.rid in
> >>>>>>      try
> >>>>>>          let fct = function_of_type ty in
> >>>>>>          let t =
> >>>>>> @@ -379,7 +382,7 @@ let process_packet ~store ~cons ~doms ~con ~tid
> >>>>>> ~rid ~ty ~data =
> >>>>>>              else
> >>>>>>                  Connection.get_transaction con tid
> >>>>>>              in
> >>>>>> -        let response = input_handle_error ~cons ~doms ~fct ~ty ~con
> >>>>>> ~t ~rid ~data in
> >>>>>> +        let response = input_handle_error ~cons ~doms ~fct ~con ~t
> >>>>>> ~req in
> >>>>>>            (* Put the response on the wire *)
> >>>>>>          send_response ty con t rid response
> >>>>>> @@ -412,11 +415,13 @@ let do_input store cons doms con =
> >>>>>>      if newpacket then (
> >>>>>>          let packet = Connection.pop_in con in
> >>>>>>          let tid, rid, ty, data = Xenbus.Xb.Packet.unpack packet in
> >>>>>> +        let req = {Packet.tid; Packet.rid; Packet.ty; Packet.data} in
> >>>>>> +
> >>>>> I think this change breaks the build with older ocaml versions:
> >>>>> 
> >>>>> root@haswell> ocamlopt -v
> >>>>> The OCaml native-code compiler, version 4.00.1
> >>>>> Standard library directory: /usr/lib64/ocaml
> >>>>> root@haswell> ocamlopt -g -ccopt "    " -dtypes -I
> >>>>> /root/tmp/xen/tools/ocaml/xenstored/../libs/xb -I
> >>>>> /root/tmp/xen/tools/ocaml/xenstored/../libs/mmap -I
> >>>>> /root/tmp/xen/tools/ocaml/xenstored/../libs/xc -I
> >>>>> /root/tmp/xen/tools/ocaml/xenstored/../libs/eventchn -cc gcc -w F
> >>>>> -warn-error F -c -o process.cmx process.ml
> >>>>> root@haswell>
> >>>>> 
> >>>>> 
> >>>>> root@ovs104> ocamlopt -v
> >>>>> The Objective Caml native-code compiler, version 3.11.2
> >>>>> Standard library directory: /usr/lib64/ocaml
> >>>>> root@ovs104> ocamlopt -g -ccopt "    " -dtypes -I
> >>>>> /root/tmp/xen/tools/ocaml/xenstored/../libs/xb -I
> >>>>> /root/tmp/xen/tools/ocaml/xenstored/../libs/mmap -I
> >>>>> /root/tmp/xen/tools/ocaml/xenstored/../libs/xc -I
> >>>>> /root/tmp/xen/tools/ocaml/xenstored/../libs/eventchn -cc gcc -w F
> >>>>> -warn-error F -c -o process.cmx process.ml
> >>>>> File "process.ml", line 487, characters 23-24:
> >>>>> Error: Syntax error
> >>>>> root@ovs104>
> >>>>> 
> >>>>> I don't know much about ocaml (OK, I know *nothing* about ocaml) so I
> >>>>> can't say what exactly might be wrong.
> >>>> Could you perhaps try this instead?
> >>>> 
> >>>> let req = {tid = Packet.tid; rid = Packet.rid; ty = Packet.ty; data =
> >>>> Packet.data} in
> >>>> 
> >>>> It is most likely that the older version of Ocaml can't infer the order
> >>>> of fields.
> >>>> 
> >>>> ~Andrew
> >>> 
> >>> 
> >>> No, it now gives me "Error: Unbound record field label tid"
> >>> 
> >>> -boris
> >> 
> >> Andrew's guess was close, but the wrong way around -- please could you try the
> >> following with the older compiler?
> >> 
> >> let req = {Packet.tid=tid; Packet.rid=rid; Packet.ty=ty; Packet.data=data} in
> >> 
> >> I was using a syntactic feature of OCaml called 'field punning' which is
> >> generally considered good practice and makes for more readable code. It looks
> >> like this feature was introduced in OCaml 3.12.0 (dating from 2010), which is
> >> consistent with Boris' findings.
> >> 
> >> What's the policy here -- is there a defined version of the OCaml compiler which
> >> tools/ocaml needs to be able to compile with?
> > 
> > It is not explicitly listed in README or INSTALL. The ocaml tools
> > maintainer (Dave in this case) is welcome to provide the minimum version
> > required.
> > 
> > Meanwhile, I don't think we should break existing build without pinning
> > down the minimum required version first, so we should fix Boris's
> > breakage.  The fix seems simple enough anyway.
> 
> It looks like the fix is small and easy — I think this is good for now.
> 
> Let’s postpone requiring a later OCaml version until we really need a feature only present in a later version. I suspect this will happen eventually, probably when we try to add a dependency (e.g. from the Mirage world) which requires 4.02+.

OK; sounds sensible.

Since the original patch has already been committed to staging, I presume you'd
like me to formally submit a standalone patch that fixes this issue. I will post
it separately.

If I'm wrong and you'd like me to post a v2 of the whole series, let me know.

Thanks,
Jonathan

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel