|
|
|
|
"readme": "# Controlling Flow: callbacks are easy\n\n## What's actually hard?\n\n- Doing a bunch of things in a specific order.\n- Knowing when stuff is done.\n- Handling failures.\n- Breaking up functionality into parts (avoid nested inline callbacks)\n\n\n## Common Mistakes\n\n- Abandoning convention and consistency.\n- Putting all callbacks inline.\n- Using libraries without grokking them.\n- Trying to make async code look sync.\n\n## Define Conventions\n\n- Two kinds of functions: *actors* take action, *callbacks* get results.\n- Essentially the continuation pattern. Resulting code *looks* similar\n to fibers, but is *much* simpler to implement.\n- Node works this way in the lowlevel APIs already, and it's very flexible.\n\n## Callbacks\n\n- Simple responders\n- Must always be prepared to handle errors, that's why it's the first argument.\n- Often inline anonymous, but not always.\n- Can trap and call other callbacks with modified data, or pass errors upwards.\n\n## Actors\n\n- Last argument is a callback.\n- If any error occurs, and can't be handled, pass it to the callback and return.\n- Must not throw. Return value ignored.\n- return x ==> return cb(null, x)\n- throw er ==> return cb(er)\n\n```javascript\n// return true if a path is either\n// a symlink or a directory.\nfunction isLinkOrDir (path, cb) {\n fs.lstat(path, function (er, s) {\n if (er) return cb(er)\n return cb(null, s.isDirectory() || s.isSymbolicLink())\n })\n}\n```\n\n# asyncMap\n\n## Usecases\n\n- I have a list of 10 files, and need to read all of them, and then continue when they're all done.\n- I have a dozen URLs, and need to fetch them all, and then continue when they're all done.\n- I have 4 connected users, and need to send a message to all of them, and then continue when that's done.\n- I have a list of n things, and I need to dosomething with all of them, in parallel, and get the results once they're all complete.\n\n\n## Solution\n\n```javascript\nvar asyncMap = require(\"slide\").asyncMap\nfunction writeFiles (files, what, cb) {\n asyncMap(files, function (f, cb) {\n fs.writeFile(f, what, cb)\n }, cb)\n}\nwriteFiles([my, file, list], \"foo\", cb)\n```\n\n# chain\n\n## Usecases\n\n- I have to do a bunch of things, in order. Get db credentials out of a file,\n read the data from the db, write that data to another file.\n- If anything fails, do not continue.\n- I still have to provide an array of functions, which is a lot of boilerplate,\n and a pita if your functions take args like\n\n```javascript\nfunction (cb) {\n blah(a, b, c, cb)\n}\n```\n\n- Results are discarded, which is a bit lame.\n- No way to branch.\n\n## Solution\n\n- reduces boilerplate by converting an array of [fn, args] to an actor\n that takes no arguments (except cb)\n- A bit like Function#bind, but tailored for our use-case.\n- bindActor(obj, \"method\", a, b, c)\n- bindActor(fn, a, b, c)\n- bindActor(obj, fn, a, b, c)\n- branching, skipping over falsey arguments\n\n```javascript\nchain([\n doThing && [thing, a, b, c]\n, isFoo && [doFoo, \"foo\"]\n, subChain && [chain, [one, two]]\n], cb)\n```\n\n- tracking results: results are stored in an optional array passed as argument,\n last result is always in results[results.length - 1].\n- treat chain.first and chain.last as placeholders for the first/last\n result up until that point.\n\n\n## Non-trivial example\n\n- Read number files in a directory\n- Add the results together\n- Ping a web service with the result\n- Write the response to a file\n- Delete the number files\n\n```javascript\nvar chain = require(\"slide\").chain\nfunction myProgram (cb) {\n var res = [], last = chain.last, first = chain.first\n chain([\n [fs, \"readdir\", \"the-directory\"]\n , [readFiles, \"the-directory\", last]\n , [sum, last]\n , [ping, \"POST\", \"example.com\", 80, \"/foo\", last]\n , [fs, \"writeFile\", \"result.txt\", last]\n , [rmFiles, \"./the-directory\", first]\n ], res, cb)\n}\n```\n\n# Conclusion: Convention Profits\n\n- Consistent API from top to bottom.\n- Sneak in at any point to inject functi
|
