这么写肯定是一大块了，一行能写下就不会这么写。
内容是什么倒是不重要。只是想稍微了解下json结构。比如有多少字，多大。这样。
或者你可以用pastebin来贴
我用 emacs-parson/bench.json at master · syohex/emacs-parson · GitHub 提供的 json 字符串来测试，是想也同时比较一下 native 跟 dynamic module。
GitHub - syohex/emacs-parson: JSON parser with dynamic module feature 展示的测试结果，除第一项之外，其它都是 dynamic module 大幅领先。而 native 的最后一项几乎与 json.el 持平。
这个 json string 只有 20 多行，比起极端样本数据，更接近 lsp 通信的数据量。在我这里，benchmark 测试结果和实际的 lsp + native 使用体验是一致的：感受不到性能提升。
用你这个测试代码和json测试文件在linux上跑了一下, 10年前的笔记本. 结果也是不太好.
Elapsed time: 31.353811s (18.417009s in 729 GCs)
Elapsed time: 42.392202s (29.418018s in 1165 GCs)
emacs编译用-O2, 测试结果显示, native json的性能仅提升了1/4, 而且时间还是在gc里. 挺尴尬啊.
把测试文件扩大了以下, 到1000行大小, 运行1000次, 测试结果:
Elapsed time: 11.953426s (6.324524s in 250 GCs)
Elapsed time: 14.364171s (9.626586s in 386 GCs)
性能提升确实不多, 怪不得真实使用起来感觉不到差别.
是挺尴尬的。
27 发布在即了吧，估计短时间不会有太大提升，等 native 还不如 dynamic module 靠谱。
看大家的测试数据，和目前的体验确实差不多。
太囧了 
这里头是出了什么问题？是Jansson的实现不好吗？还是说Elisp绑定没做好？
我在mac上用MS的word软件编辑doc时，会出现拷贝快捷键Command＋C失效的情况，有时候连按两次也失效，所以现在每次拷贝就连按4下！
When performing benchmark with native parser wrap the call to the native parsing with (with-temp-buffer …) - related discussion - bug#31138: Native json slower than json.el and [discussion] writing json parser as rust module · Issue #4 · ubolonton/emacs-module-rs · GitHub
1 个赞
I will post the results with elp performance testing and with enabled both native parsing + json el to have comparison in the real usecase: (compare lsp--read-json-el and lsp--read-native-json the code available at Use native JSON functionality in Emacs 27 · Issue #210 · emacs-lsp/lsp-mode · GitHub )
lsp--parser-on-message 171 4.825922573 0.0282217694
lsp--read-json 171 4.601956367 0.0269120255
lsp--read-json-el 171 3.9113452219 0.0228733638
lsp--parser-read 1307 1.0518447099 0.0008047778
lsp--read-native-json 171 0.688050226 0.0040236855
lsp--on-notification 130 0.0726324090 0.0005587108
lsp--on-diagnostics 7 0.063708947 0.0091012781
lsp-ui-sideline 30 0.05643676 0.0018812253
lsp--make-message 56 0.0562572890 0.0010045944
lsp-ui-sideline--diagnostics-changed 7 0.055627562 0.0079467945
lsp--send-request-async 43 0.0467603880 0.0010874508
lsp-ui-doc--on-hover 2 0.029292634 0.014646317
lsp-ui-sideline--run 7 0.026202328 0.0037431897
1 个赞
都不想升级到27。。。
The (with-temp-buffer ...) does not make native parsing faster:
⋊> ./src/emacs -Q --batch -l ~/.scratch/emacs/benchmark-native-json.el
(w/benchmark '100000 '(json-parse-string json-string)) ;; => Elapsed time: 25.451464s (16.632284s in 728 GCs)
(w/benchmark '100000 '(json-read-from-string json-string)) ;; => Elapsed time: 30.148163s (18.175661s in 866 GCs)
⋊> ./src/emacs -Q --batch -l ~/.scratch/emacs/benchmark-native-json.el
(w/benchmark '100000 '(json-parse-string json-string)) ;; => Elapsed time: 22.577644s (14.639583s in 728 GCs)
(w/benchmark '100000 '(json-read-from-string json-string)) ;; => Elapsed time: 38.248179s (22.567974s in 866 GCs)
⋊> ./src/emacs -Q --batch -l ~/.scratch/emacs/benchmark-native-json.el
(w/benchmark '100000 '(json-parse-string json-string)) ;; => Elapsed time: 22.662395s (14.645539s in 728 GCs)
(w/benchmark '100000 '(json-read-from-string json-string)) ;; => Elapsed time: 29.729253s (17.643603s in 866 GCs)
⋊> ./src/emacs -Q --batch -l ~/.scratch/emacs/benchmark-native-json.el
(w/benchmark '100000 '(json-parse-string json-string)) ;; => Elapsed time: 22.388149s (14.668245s in 728 GCs)
(w/benchmark '100000 '(json-read-from-string json-string)) ;; => Elapsed time: 29.638171s (17.242109s in 866 GCs)
⋊> ./src/emacs -Q --batch -l ~/.scratch/emacs/benchmark-native-json.el
(w/benchmark '100000 '(json-parse-string json-string)) ;; => Elapsed time: 27.819788s (15.239697s in 728 GCs)
(w/benchmark '100000 '(json-read-from-string json-string)) ;; => Elapsed time: 30.269438s (18.141456s in 858 GCs)
⋊> ./src/emacs -Q --batch -l ~/.scratch/emacs/benchmark-native-json.el
(w/benchmark '100000 '(json-parse-string json-string)) ;; => Elapsed time: 26.836383s (14.706035s in 728 GCs)
(w/benchmark '100000 '(json-read-from-string json-string)) ;; => Elapsed time: 29.445384s (17.515654s in 858 GCs)
I have updated Emacs to the recent version of the master branch.
Here is my test code: benchmark native and json.el #emacs · GitHub
I mentioned in the lsp-mode issue, there is no much effect in small jsons since I believe that the elisp funcall is as slow as the parsing itself.
Here are the results with large.json · GitHub (the first run is without (with-temp-buffer ...)
➜ Desktop ~/.bin/emacs-latest/bin/emacs -Q --batch -l benchmark-native-json.el
GNU Emacs 27.0.50 (build 7, x86_64-pc-linux-gnu, GTK+ Version 3.22.30) of 2019-03-10
(benchmark 100 '(json-read-from-string json-string)) ;; => Elapsed time: 18.488213s (10.891228s in 1365 GCs)
(benchmark 100 '(json-parse-string json-string)) ;; => Elapsed time: 8.127364s (1.391082s in 100 GCs)
➜ Desktop ~/.bin/emacs-latest/bin/emacs -Q --batch -l benchmark-native-json.el
GNU Emacs 27.0.50 (build 7, x86_64-pc-linux-gnu, GTK+ Version 3.22.30) of 2019-03-10
(benchmark 100 '(json-read-from-string json-string)) ;; => Elapsed time: 18.890727s (11.152488s in 1365 GCs)
(benchmark 100 '(json-parse-string json-string)) ;; => Elapsed time: 5.538966s (1.364898s in 100 GCs)
In my experiencing, lsp-mode has more features than eglot, but it’s slower. Looks like the bottleneck is parsing JSON. What’s the difference of parsing JSON between lsp-mode and eglot? And the doc UI of eglot looks nicer than lsp-mode now although it didn’t.
There was one extremely slow piece of code on the hot path (see lsp--sort-completions in Native json support by yyoncho · Pull Request #691 · emacs-lsp/lsp-mode · GitHub) . In addition to that, the native json will speedup lsp-mode too. After these changes, on my 4 years old PC lsp-mode feels pretty much instant and visually there is no speed difference between lsp-mode and vscode(except maybe dart language server which returns 3.5 mb results). I would point also that the completion filtering should be moved on the client. I think that @tigersoldier is working on that and this will improve the experience a lot.
Can you post screenshots about the doc(or directly a PR)? I must admit I haven’t spent a lot of time working on lsp-ui(although there is a lot of stuff to be done). The priority for now is to cover the lsp spec so we could use that as a base for further improvements.
If you see an area in which lsp-mode is lagging behind any other language server client just go ahead and file a PR.
Vibhab, the author of lsp-mode is also returning which will improve the pace a bit. I am spending ~10 hours a week coding which is not enough to cover all the things that I want to do in the packages that I am interested in (lsp-mode/dap-mode/lsp-ui/lsp-ide/helm-lsp/spacemacs/lsp-ide/lsp-treemacs/lsp-java) but still I believe that lsp-mode has improved a lot since 3-4 months ago. Still given the scope of the project lsp-mode ecosystem needs more contributors/members/maitainers.
6 个赞
Thank you for the update! I am impressive of the improvements of lsp-mode recently. I believe it has bight feature. I am using lsp-mode with pls and bingo all the time, but sometimes try eglot and do some comparisons. The screenshots are below.
- eglot
- lsp-mode
FYI eglot uses gfm.el
Yes， seems it’s rendered as markdown buffer.
I recalled it from my memory. I just looked up the source, seems eglot uses gfm-view-mode in markdown.el (instead of gfm.el, which doesn’t exist).