Hello again, Dmitry.

I managed to produce a benchmark with the following JSON file (560K,
~10,000 lines): http://folk.uio.no/simenheg/huge.json.

I read it into `huge-json', and ran the following before the patch:

(benchmark-run 100 (json-read-from-string huge-json))
     ⇒ (19.421527195 938 7.448190372000017)

And after the patch:

(benchmark-run 100 (json-read-from-string huge-json))
     ⇒ (19.321667537 997 7.218977015999971)

-- Simen

On Sun, Nov 1, 2015 at 8:52 PM, Simen Heggestøyl <simenheg@gmail.com> 
wrote:
> Hi Dmitry, thanks for the feedback!
> 
> On Sat, Oct 31, 2015 at 3:23 PM, Dmitry Gutov <dgutov@yandex.ru> 
> wrote:
>> The video looks great, but the inline patch lacks indentation, and 
>> doesn't apply (dunno if the former is the cause of the latter). 
>> Please resend it as an attachment.
> 
> Ah, yes, it appear that the whitespace got lost in the email for some
> reason. I'll try attaching it to this email.
> 
> To test it, I've been using the following interactive function:
> 
>   (defun json-mode-show-path ()
>     "Show the path to the JSON value under point."
>     (interactive)
>     (let ((path (json-path-to-position (point))))
>       (if path
>           (let ((formatted-path
>                  (json-mode--format-path (plist-get path :path))))
>             (pulse-momentary-highlight-region
>              (plist-get path :match-start)
>              (plist-get path :match-end))
>             (message formatted-path)
>             (kill-new formatted-path))
>         (message "Not a JSON value"))))
> 
>   (defun json-mode--format-path (path)
>     "Return PATH formatted as a JSON data selector.
>   PATH should be a list of keys, which can be either strings or
>   integers."
>     (mapconcat (lambda (key) (format "[%S]" key)) path ""))
> 
> It's also included it in json-mode.el:
> http://folk.uio.no/simenheg/json-mode.el.
> 
>> Without trying it, my main concern would be any performance 
>> regression to json-read (it's not particularly fast already). Have 
>> you done any benchmarking?
> 
> I agree that would be bad. I'll try to produce some benchmarks when
> I've got some free time on my hands!
> 
> -- Simen