I am trying to copy the Erlang idiom of "calling" another process through a
function call interface. That is, suppose I have two named processes
corresponding to modules `caller` and `callee`
caller.erl:
foo() ->
Result = callee:get_some_data(),
do_something_with(Result).
callee.erl:
get_some_data() ->
gen_server:call(?MODULE, get_some_data)
handle_call(get_some_data, CalleeState) ->
{Result, NewState} = do_get_some_data(CalleeState),
{reply, Result, NewState}.
I'm trying to replicate that in JS, because it's definitely the most pleasant
idiom for having different processes "talk" to each other. This also allows
the callee process to define an API that black-boxes away the internal
messaging structure, thus giving the developer the flexibility to change that
internal messaging structure without breaking the calling code.
======================
NACL mental disability
======================
NACL is mentally disabled and can't be loaded as an ES6 module. That is, it can't be
"imported" using
```js
import * as nacl from './path/to/nacl.js';
```
Instead you have to include it in a page. So we have a page called
`background.html`.
I naively tried this
```html
<script src="dist/jex_include/local-nacl-1.0.3/nacl.js"></script>
```
That of course didn't work because we live in hell.
I initially thought the problem was an invalid hash. No. Notes as I figure it
out:
Trying to figure out why TweetNaCl isn't loading
ok so the problem is not an invalid hash, it's something specific
to nacl
to have an inline script, its hash must be specified in
manifest.json
example (integrity key is not necessary, it is there purely for
comment purposes)
beginning of line
V
| <script integrity="sha256-8T3dvQxHPNQvgxzCemw3KGUCM5RhknpCQF6pCcTdFTw=">console.log('bg-test-inline.js line 1!');
|</script>
newline is there for hash purposes
this works if and only if
'sha256-8T3dvQxHPNQvgxzCemw3KGUCM5RhknpCQF6pCcTdFTw=' is included
in the manifest.json
example (top level key):
"content_security_policy" : "script-src 'self' 'sha256-8T3dvQxHPNQvgxzCemw3KGUCM5RhknpCQF6pCcTdFTw='; object-src 'self'",
Absent that explicit allow, I get this error from Firefox:
Content Security Policy: The page’s settings blocked the loading of a resource at inline (“script-src”).
Refs:
1. https://developer.mozilla.org/en-US/docs/Mozilla/Add-ons/WebExtensions/manifest.json/content_security_policy
2. https://developer.mozilla.org/en-US/docs/Mozilla/Add-ons/WebExtensions/Content_Security_Policy#inline_javascript
If I include the same script as a file,
<script src="bg-test-inline.js"></script>
this works irrespective of if I specify it as allowable in the manifest
nacl is just refusing to load
Loading failed for the <script> with source “moz-extension://eafd6a74-b75b-4312-9a34-8087b68395e7/dist/jex_include/local-nacl-1.0.3/nacl.js”.
I initially thought it was a hash problem, but it is not
TweetNaCL doesn't work as an ES6 module, it works by destructively
updating the window namespace. This I think was necessary before
ES6, and that behavior is retained either because of laziness or
because of backward compatibility reasons.
hmm
so this just plainly doesn't work and i cannot figure out why
<script src="dist/jex_include/local-nacl-1.0.3/nacl.js"></script>
I think it works if I do it in the popup window instead. Let's try
oh no
no
oh my god
i am so stupid
the problem was that i typed the url wrong
the package is "tweetnacl" not "nacl", and i also need /dist/ too
<script src="dist/jex_include/local-tweetnacl-1.0.3/dist/nacl.js"></script>
that works
all this time i thought that it was the browser that was mentally disabled
it turned out the mental disability was inside me all along
alright well I discovered a potential future pitfall at least
Methodology:
- in Erlang, generate random DecodeData that are decentish (hard to generate
because of arbitrary depth recursive nature)
- in Erlang, check that DecodeData->rlp:encode->rlp:decode is an identity (it
is!)
- generate tons of encode/decode pairs for Python, make sure they all work
(they do!)
my current methodology for doing X
- find library that does X in Python with lots of stars on GitHub
- autogenerate test cases against Python library
- do X in Erlang to work out logic in easier setting
- make sure Erlang implementation works
- do X in TypeScript
- make sure TypeScript implementation works
It compiles but there's some NYI functions
This is going to have all the common functions that we need all over the place
- base64 encode/decode
- base58 encode/decode
- rlp encode/decode
- transaction types
- transaction formation
- reading stuff from the transaction
- safe type
- (maybe) AWCP
This commit contains a refactored version of my prior base64 implementation,
which is more imperativey and less "trying to fit Erlang into TS".
pharpend