Working on -e trace

[Dmitry V. Levin]

On Thu, Mar 10, 2016 at 11:42:24PM +0530, haris iqbal wrote:
> On Thu, Mar 10, 2016 at 5:24 AM, Dmitry V. Levin wrote:
> > On Wed, Mar 09, 2016 at 11:02:52AM +0530, haris iqbal wrote:
> > [...]
> >> Sorry for the multiple commits. My vim is giving trouble.
> >
> > Please don't send incremental commits, just amend and send the one you
> > already have.
> 
> I did not quite understand this. So, I make changes and send the older
> commit log? How will the changes be seen by you guys then?

Could you submit a patch implementing the change as a single commit,
not as a series of commits, please?

> >> >> --- a/syscall.c
> >> >> +++ b/syscall.c
> >> >> @@ -78,6 +78,7 @@
> >> >>  #define TP TRACE_PROCESS
> >> >>  #define TS TRACE_SIGNAL
> >> >>  #define TM TRACE_MEMORY
> >> >> +#define TSC TRACE_SCHED
> >> >>  #define NF SYSCALL_NEVER_FAILS
> >> >>  #define MA MAX_ARGS
> >> >>  #define SI STACKTRACE_INVALIDATE_CACHE
> >> >> @@ -500,6 +501,8 @@ lookup_class(const char *s)
> >> >>          return TRACE_DESC;
> >> >>      if (strcmp(s, "memory") == 0)
> >> >>          return TRACE_MEMORY;
> >> >> +    if (strcmp(s, "scheduler") == 0)
> >> >> +        return TRACE_SCHED;
> >> >>      return -1;
> >> >>  }
> >> >
> >> > Please follow the indentation style.
> >> > BTW, why "scheduler"?  Could you give a rationale for this name?
> >>
> >> I thought as all the system calls are one or the other related to the
> >> scheduler. We can also consider the description "syscalls relate to
> >> scheduling.
> >
> > There is a subtle difference between a scheduler (an actor) and a set
> > of syscalls related to scheduling (an action) than might lead to confusion
> > in the future.  That's why I still think that "%sched" would be better in
> > long run than "scheduler" or "scheduling".
> 
> True, but my concern was we cannot change the name of this new one to
> "%sched", while the older ones are still following some other naming
> convention. What can be done is, we can decide on the names for all
> the existing options (for file, memory, ..) and then change all in one
> commit.

All currently supported names of syscall sets will remain supported no
matter what kind of naming scheme will be used later.  So once we added
"scheduler" to the user API, we would have to keep it.

> --- a/NEWS
> +++ b/NEWS
> @@ -12,6 +12,7 @@ Noteworthy changes in release ?.?? (????-??-??)
>      spent in decoding socket details in -yy mode.
>    * Implemented decoding of copy_file_range syscall.
>    * Implemented dumping of vmsplice syscall.
> +  * Added feature to trace scheduler related system calls with option -e.

Please change the wording to follow the style.  You can use
"-e trace=memory" in the same file as an example.

> --- a/strace.1
> +++ b/strace.1
> @@ -423,6 +423,9 @@ Trace all file descriptor related system calls.
>  .BR "\-e\ trace" = memory
>  Trace all memory mapping related system calls.
>  .TP
> +.BR "\-e\ trace" = scheduler
> +Trace all scheduler related system calls.
> +.TP
>  \fB\-e\ abbrev\fR=\,\fIset\fR
>  Abbreviate the output from printing each member of large structures.
>  The default is

That's fine.


-- 
ldv
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.strace.io/pipermail/strace-devel/attachments/20160310/a34ad2f9/attachment.bin>