From 35a54abc319a76f39ebbbc245b14a655f492dc29 Mon Sep 17 00:00:00 2001
From: charguer <arthur@chargueraud.org>
Date: Wed, 2 Mar 2016 15:10:46 +0100
Subject: [PATCH] todo_file
---
generator/TODO | 163 ++++++++++++++++++++++++++++++++-----------------
1 file changed, 107 insertions(+), 56 deletions(-)
diff --git a/generator/TODO b/generator/TODO
index 7981a9b..47c054c 100644
--- a/generator/TODO
+++ b/generator/TODO
@@ -1,96 +1,147 @@
-- clean up stdlib.ml
-- implement stdlib.js to match all functions of stdlib.ml
-- remove JsNumber by binding a set of operations in stdlib.js
- implemented using builtin JS primitives
-- reactivate
- default: throw "No matching case for switch";
- by having js_of_pattern return an additional boolean indicating
- whether there was a default case or not.
-
-- optimize the lineof file to generate code that, at initialization
- fills in an array with brute data, and a function that performs
- look ups in the array to get the output
+=========================================================
+FIRST
+
+- remove intermediate folder "miniml" in "../jsjsref/miniml/.."
-- update the code that translates esprima syntax to recognize
- "use strict" directives, which come as first statement of
- a body, in hte form of a raw expression of raw type "use script".
- => TODO: look for code that already does this for esprima.
+=========================================================
+ARTHUR
-- check in Shared.ml whether we need to take absolute values
- let string_sub s n l =
- substring (int_abs n) (int_abs l) s
+- reimplement "lineof" using a table
-- here:
- string_sub str (int_of_float k0) 1
- => TODO: check k0 is not negative
+- generate the escaped source code
+- control which file should feature logging
=========================================================
+NEXT
+- introduce testing functions called
+ - test_lineof("filename", "token") -> returns a "pos" object
+ - test_parse("code") -> log esprima parse and log convert
+ - test_exec("code") -> parse, convert, build trace, log trace
-LATER
-- cases seem to be in reversed order, check and fix
- => corriger à la main le jour où on s'amuse à relire le code.
- => order in pattern matching extraction seems to follow
- order from the definition of the inductive type,
- instead of the order of the code.
+- clean up stdlib.ml
+ - remove unused functions
+ - sort out remaining ones
-LATER
-- untab closing bracket for fun def
- => later maybe
+- implement stdlib.js
+ - need to have one function for each one of stdlib.ml
+ - number related functions are implemented with JS counterpart directly
+- display JS values/env/heap/ctx in html
+ - contexts: essentially as before
+ - environments: need to follow the list of environment records
+ - heap: same as before, and do not attempt at first to show
+ additional information in the heap appart from key/value bindings.
+- fill in the "esprima-to-ast.js" file, following the template
+ (throw an exception, caught and displayed as an alert message,
+ in case the source program uses for-in).
-- Choose a small example
+- test langage features one by one, using minimal input programs;
+ these programs should be stored in an "examples/" folder;
+ for each example, check the AST, check step-by-step the execution.
+- test a compound example, e.g.:
var x = 3;
x = 4;
if (x === 4) {
while (true) {}
}
-- write a JS function that takes an esprima AST as input (sub-language for the
- moment) that generates an AST in our syntax (encoded of concrete types):
- { type : constructor; label0 : value ; … }
- this should remember locations
-- edit JsSyntax.ml to put readable labels
+=========================================================
+LATER
-- write a driver in JS
- - take a program in the text area
- - parse to esprima
- - translate to our syntax
- - call JsInterpreter.js using the correct function (see run_js.ml)
+- update the code that translates esprima syntax to recognize
+ "use strict" directives, which come as first statement of
+ a body, in hte form of a raw expression of raw type "use script".
+ (look on the web for code that already does this for esprima).
+
+- set up the call to the appropriate function for propagating the
+ "strictness" info in the AST.
+
+- add a combo box above the source textbox for loading in one click
+ a particular example. The source code for the examples is obtained
+ by including a JS file that contains an array of source code,
+ this file being generated from the contents of the "examples" folder.
+
+- put online the tool on the jscert website
+ - make sure to describe the supported features
+ (ecma5 minus for-in and a few other things)
+ - explain that currently JS files are generated
+ from ML files and thus should not be edited,
+ although this might change in the future.
+
+- full display of heap information
+ add a "+details" button
+ to show additional information about an object, like the
+ getter and setter methods, and also for each field we need
+ to see additional information like enumerable/configurable...
+
+- search by predicate in the trace
+ - remove the "with (obj)" hack that is currently there
+ - introduce functions
+ - INTERP_RAW("x") : returns the value bound to "x" in the
+ current context, by looking up the linked list of arrays
+ - PROG_RAW("x") : returns the value bound to "x" in the
+ environment, using the prototype chain resolution function
+ - JSOF(v) : lift the encoding of a JS value to the corresponding JS value
+ - INTERP_VAL("x") = JSOF(INTERP_RAW("x"))
+ - PROG_VAL("x") = JSOF(PROG_RAW("x"))
+
+- Reactivate the generation of the lines of the form:
+ default: throw "No matching case for switch";
+ by having js_of_pattern return an additional boolean
+ indicating whether there was a default case or not.
+ This is needed to detect bugs in the spec where the
+ switch are missing useful cases.
-- translate stdlib.ml into stdlib.js
-- correct the Makefile to make sure we do not generate everything all the time
+=========================================================
+LATER LATER
+- There is one place where we compute a "substring" in the code;
+ need to check whether the argument is always nonnegative.
+- Coq extraction has generated cases in the order of the definition
+ of the constructors in the inductive definition; this is not the
+ same order as in the original coq files; we should put back in
+ the right order, i'm afraid we need to do this by hand.
-- => binder qq part vers une comparaison --- dans un fichier js
- _compare_JsSyntax_native_error
- _compare_JsSyntax_prealloc
+- Investigate generation of english sentences from esprima AST
+ of the code of the interpreter, with live visualization of
+ the sentences being executed.
+=========================================================
+OPTIMIZATIONS
+
+- The representation of the heap could be optimized by
+ having "remove" nodes in the list describing a heap,
+ this would avoid filtering through the heap in O(n).
-==========
+- Optimize heap representations using trees, to improve
+ lookups from O(n) to O(log n).
-TODO: demo of double clicking goes to the right page in ecma in english
+- Only introduce logging instructions in files for which
+ they are going to be used.
-TODO: interpretation of comparison function in the search by "predicate"
- might want to automatically "lift" basic values to their encoding
- (unless some box is checked to prevent this behavior)
- example: "x == 3" means:
- "similar(x, { type: "value", tag: "value_prim", value: { type: "prim", tag: "float", value: 3.0 } })".
+=========================================================
+NEVER
+
+- There is an extra tab when closing bracket for fun def
+
+
+
+
+=========================================================
- where similar is a recursive structural comparison function (not going through the heap)
-==========
--
GitLab