Tuesday, June 16, 2020

Profiling vim start up time

After updating my vim plugins and adding a few new ones, I found my start up time became unreasonably slow. I blamed my tag generating plugin because of how the slowdown only occurred when opening vim with a file. That was a red herring, removing it didn't speed things up. Checking that projects issues I found that it was already switched to be asynchronous.

While investigating, I found a reference to the --startuptime option in vim. This option creates a log file containing entries for each startup action: sourcing files, executing plugins etc with timing information showing start and end times for each action, measuring relatively from the start and from the previous command. It's a lot of useful information for investigating which plugins are slowing startup time, reported in a useful format. Thumbs up to vim1

% vim --startuptime /tmp/startup-time.log
% vim --startuptime /tmp/startup-time-vimrc.log ~/.vimrc

Wow, loading all these new plugins was taking a while. 800ms in the first case but 3200ms when loading a file! More files are sourced in the latter case as the FileType files need to be sourced and applied.


starting time without a file load

471.134  121.649: VimEnter autocommands
471.137  000.003: before starting main loop
477.200  004.424  004.424: sourcing /home/andrew/.vim/bundle/vim-fugitive/autoload/fugitive.vim
481.049  002.378  002.378: sourcing /home/andrew/.vim/bundle/tagbar/autoload/tagbar.vim
481.569  000.173  000.173: sourcing /home/andrew/.vim/bundle/tagbar/autoload/tagbar/debug.vim
851.768  000.774  000.774: sourcing /home/andrew/.vim/bundle/tagbar/autoload/tagbar/types/uctags.vim
852.226  000.177  000.177: sourcing /home/andrew/.vim/bundle/tagbar/autoload/tagbar/prototypes/typeinfo.vim
861.475  382.412: first screen update
861.476  000.001: --- VIM STARTED ---

Starting time loading .vimrc

585.708  181.604: VimEnter autocommands
585.715  000.007: before starting main loop
595.217  000.623  000.623: sourcing /home/andrew/.vim/bundle/vim-airline/autoload/airline/async.vim
606.194  006.914  006.914: sourcing /home/andrew/.vim/bundle/tagbar/autoload/tagbar.vim
606.877  000.122  000.122: sourcing /home/andrew/.vim/bundle/tagbar/autoload/tagbar/debug.vim
3057.382  001.216  001.216: sourcing /home/andrew/.vim/bundle/tagbar/autoload/tagbar/types/uctags.vim
3057.895  000.191  000.191: sourcing /home/andrew/.vim/bundle/tagbar/autoload/tagbar/prototypes/typeinfo.vim
3065.474  000.258  000.258: sourcing /home/andrew/.vim/bundle/tagbar/autoload/tagbar/prototypes/fileinfo.vim
3242.091  000.213  000.213: sourcing /home/andrew/.vim/bundle/tagbar/autoload/tagbar/state.vim
3249.747  2655.495: first screen update
3249.750  000.003: --- VIM STARTED ---

The big difference: missing time between loading tagbar/debug and tagbar/types/uctags.vim, 370ms vs 2400ms. The files are getting sourced quickly, but something is causing an external delay.

So maybe it is the tags generating plugin locking the tags file? But a tags file neither exists nor is generated in the directory containing .vimrc. Very curious indeed.

The debug file (tagbar/debug) file defines several function! functions, and has one minor branch on if has('realtime'). Nothing suspicious at all.
Similarly, the uctags file (tagbar/types/uctags.vim) creates a single function function! tagbar#types#uctags#init(supported_types) abort which doesn't seem to be affected by the currently loaded file.

480.725  101.452: VimEnter autocommands
480.732  000.007: before starting main loop
490.939  000.468  000.468: sourcing /home/andrew/.vim/bundle/vim-airline/autoload/airline/async.vim
498.378  003.465  003.465: sourcing /home/andrew/.vim/bundle/tagbar/autoload/tagbar.vim
499.003  000.131  000.131: sourcing /home/andrew/.vim/bundle/tagbar/autoload/tagbar/debug.vim
2947.025  000.461  000.461: sourcing /home/andrew/.vim/bundle/tagbar/autoload/tagbar/types/uctags.vim
2947.325  000.041  000.041: sourcing /home/andrew/.vim/bundle/tagbar/autoload/tagbar/prototypes/typeinfo.vim
2954.640  000.096  000.096: sourcing /home/andrew/.vim/bundle/tagbar/autoload/tagbar/prototypes/fileinfo.vim
3179.663  000.203  000.203: sourcing /home/andrew/.vim/bundle/tagbar/autoload/tagbar/state.vim
3188.567  2703.970: first screen update
3188.572  000.005: --- VIM STARTED ---

330.198  006.165: BufEnter autocommands
330.201  000.003: editing files in windows
330.967  000.694  000.694: sourcing /home/andrew/.vim/bundle/YouCompleteMe/autoload/youcompleteme.vim
442.792  111.897: VimEnter autocommands
442.797  000.005: before starting main loop
449.255  000.333  000.333: sourcing /home/andrew/.vim/bundle/vim-airline/autoload/airline/async.vim
462.352  019.222: first screen update
462.358  000.006: --- VIM STARTED ---


Disabling the tagbar plugin gets me back my missing 2.5 seconds.

Tagbar has Issue #477 open for this delay during startup, it is running ctags synchronously under the hood on startup.

Further debugging

The syntastic and YouCompleteMe plugins each add ~200ms to startup time. Removing all three lowers my start time to 180ms which feels fast, as there is no noticable lag after the first paint of the screen.

I really enjoy the benefits of live linting with syntastic. Moving to Ale, an asynchoronus linter speeds up startup considerably with a ~30ms load time. I'll switch for a while and see how it compares.

Monday, June 1, 2020

Updating syntax highlighting imports to CDN (alexgorbatchev)



Syntax highlighting

For many years I have used a javascript based syntax highlighter on this blog. Including it directly from the author Alex Gorbatchev's site alexgorbatchev.com. Lately his site has been timing out behind Cloudflare (522) and slowing my initial render.

I've fixed the links to point to the wonderful javascript mirror Cloudflare provides at https://cdnjs.cloudflare.com/ajax/libs. This should fix the loading issue. I also took this opportunity to move to https loading for the embed, to fix any mixed-content warnings. Apologizes to any patron who was inconvenienced.

The Cloudflare page for this packages is https://cdnjs.com/libraries/SyntaxHighlighter, which shows all of the available files within the package.

A three-year-old answer on this lowly rated stackoverflow question provided details on the new link location.

An expansive view of back-end engineering

Originally written for an introduction email/cover letter, this ran well over blog length. I moved it here and aggressively edited the email. The email is still way too long. I get loquacious when I'm passionate and excited.
Je n’ai fait celle-ci plus longue que parce que je n’ai pas eu le loisir de la faire plus courte.
-- Blaise Pascal

I didn’t have time to write you a short letter, so I wrote you a long one.
-- Mark Twain

I, Engineer:

I am an excellent back-end Sr. Engineer experienced in a range of start-up sizes and domains. I enjoy digging into problems of scale and expanding a working system for the next 10x.

Back-end engineering requires systemic thinking and knowledge across a wide spectrum. These range from system design to production facing development to monitoring and metrics for support to data extraction and analysis to ops. I understand the big picture and also love to dig into the technical details and implementation. I build on my background as an Electrical Engineer and am happy to discuss anything in the the stack down to the metal ... oxide in the gates of the chips.

I am comfortable and extensively experienced with all the pieces of a well run system:

  • micro and macro service development including analysis of effectiveness and efficiency. A polyglot, most recently using Python.
  • Extracting loosely coupled services from a monolith.
  • metrics, monitoring and alerts to track and improve the health of the system.
  • RDBMS schema design with table and query optimization.
  • Trade offs in design and implementation of nosql cache and secondary stores.
  • Code lifetime tracking and shared ownership and support.
  • Extracting and exporting data, reporting, building a pipeline into reporting stores from OLAP to data lakes.
  • ops from provisioning raw metal to EC2 to docker image and containers.

I care deeply about removing accidental complexity and the resulting friction. Improving overall velocity and trust in the system through investments in tests, logs, monitoring and metrics. Instituting shared code ownership through testing, reviews, documentation and knowledge transfer with clear paths and branching from development, staging, QA, and production. Building a culture of continuous learning and exploration, where we all improve.

I was asked last week, "What is my dev super power?" In the moment, I said, "modesty," which is quite a pivot from the required interview energy of "I AM SO GREAT!" to humility and empathy. I listen and learn. I am kind and effusive when people come to me with questions and work together to find answers, and show how to seek further answers. I want to help everyone around me so we can all do more. I earn respect from my peers both through technical excellence and deep domain knowledge but also from consistency of character, respect and care. I invite you to reach out to my friends and coworkers to hear their opinions of my personality and character, ask for an anecdote of working together formally or informally.

There are so many "I" statements here. I am more interested in "we." I join early stage start-ups and small companies where I have the leverage to influence and expand "we." I excel at intra-team and inter-team communication, at finding and filling the cracks in the path from customer need to customer success. Building "we" in dev means realizing it is "we" not "they" with product, qa and ops, management, executives and the customer. Camaraderie is built upon mentoring, training and learning, shared purpose and unified direction. "We" also includes sharing with the community outside our walls. This is why I run and attend meet-ups, speak and conferences and volunteer with open source. This is an responsibility and obligation as a senior member of the community, to reach back and help more people climb into our club

I have the underlying fundamental skills in linux, git, apache/nginx, wsgi, postgres, mongo, elastic, memcache, redis, riak, rabbitmq and kafka, and a wide array of amazon: EC2, S3, RDS, EMR, dynamodb, sqs/sns/kinesis, redshift, athena, and etc.

A pleasure to make your acquaintance.



blog: lowlevelmanager.com splash: rockstär.dev

Friday, May 1, 2020

Rust playground from your own gist

The rust playground is awesome.

It allows playing with rust in the browser, without needing to install anything locally. These playgrounds / playpens are popular with newer languages. I've definitely seen with rust, go, kotlin, and others. Javascript ones allow script and css and other elements.

When editing a playground, the file can be modified, compiled, and run. The file can be exported as a direct link with the code (if it is short enough to fit in a get param). The code is also stored as a github gist under the "rust-play" user, and export links are provided to view the gist and to load a permalink to the playpen with that gist.

I haven't seen this documented anywhere, but we can use our own gist in the link!

Permalink looks like this:


version, mode and edition options configure the settings for the playground.
The gist option specified the gist to load.

Advantages to using your own gist:
  • Ownership and control
  • Notification of comments
  • Tracking of forks
  • Updates of gist are reflected in the playground
  • Edits made in the playground are lost.  The playground doesn't have update access to the gist.
  • Unexpected experience for consumers?

Try out my Playground link for gist 23727a722bff54fd20f44ef43b96466b

Embedding a single file from a github gist

Embedding a github gist

Github has improved their gists by directly providing an embed script.

This is pretty handy. From the sharing dropdown on the gist, choose embed and then copy the script tag. Paste that directly into your html (say on your super minimal blogging platform, like blogger/blogspot).

The script will embed the gist with syntax highlighting and links to view the raw gist.

Format of the script tag:
<script src="https://gist.github.com/${username}/${gist_id}.js"></script>

username is optional and can be skipped:
<script src="https://gist.github.com/${gist_id}.js"></script>

Embedding a single file from a gist

When our gist contains multiple files, the default embed will include all of the files. (Sorted by filename, just like in the gist). You can embed just a single file by adding the filename to a file get param.

Format of the script tag:
<script src="https://gist.github.com/${username}/${gist_id}.js?file=${filename}"></script>

The username path can be skipped:

<script src="https://gist.github.com/${gist_id}.js?file=${filename}"></script>

Embedded file example

Embedding from file is_valid_sequence.rs from my gist 23727a722bff54fd20f44ef43b96466b

<script src="https://gist.github.com/23727a722bff54fd20f44ef43b96466b.js?file=is_valid_sequence.rs"></script> 

toy interiew problems in rust

2019 was my "Year of Rust".  I'd hoped to blog about that, yet here we are. :)
 Recently I've been doing random problems at random "coding-interview" sites.  This is a bit like working musical scales. Practice in flexing the low level muscle memory to allow it to fade into the background and allow thinking at a higher level.

  I mostly do them in python to work out the kinks in my algorithm (seriously, they are all about fixating on annoying edge cases.)  And then a few I work out in rust.


Today's challenge: Check If a String Is a Valid Sequence from Root to Leaves Path in a Binary Tree on leetcode.com

Check If a String Is a Valid Sequence from Root to Leaves Path in a Binary Tree

Given a binary tree where each path going from the root to any leaf form a valid sequence, check if a given string is a valid sequence in such binary tree.
We get the given string from the concatenation of an array of integers arr and the concatenation of all values of the nodes along a path results in a sequence in the given binary tree.


A few things to note about my solution:
  • The recursion is pretty simple and makes lovely use of the match syntax
  • rust match syntax doesn't allow matching a slice into (head, tail) as one might be used to in other pattern match languages or a lisp where all arrays are built up from recursively defined two element lists, e.g. (a, (b, (c, (d, nil))))
  • During development, I worked out all the edge case logic with the recursive calls commented out and short circuited. After getting that all working, I dug into the type issues with my intial naive attempt at executing the recursion.
  • extracting a value from a Option<RC<RefCell<T>>> requires jumping through a couple of hoops.
  • My internal recursive function has a slightly different signature than the public function. In my python solution, I was able to use the primary interface in my recursion code. Passing literal vecs as borrowed from the RefCell was just never going to work.


Now let's add some tests!
The leetcode interface handles converting tests from a text form, populating the tree and then running the algorithm. Details of their population routine are not available. Adding tests manually is ... annoying. Which makes it a good practice exercise for building RC

Rust Playpen

Full code available to edit and play with on the rust playpen