Bug in compiling lambda default argument values

Transcrypt has a bug in the way it compiles code for default values of arguments to lambda functions. The bug is serious, because the ability to close on several different values in callback functions for ajax calls is crucial for ajax programming.

Python closures capture variables, not their values. The hack in Python to create a closure that captures the value of a variable rather than the variable itself is like this:

for r in rows.indexes():
    row = rows.row(r)
    deferreds.append(JQ.post("assign_to_run", {'run': run, 'sample_data': JSON.stringify([row.data()[12]])}) \
         .done(lambda *, rw=row: rw.deselect() ))

but Transcrypt does not capture the value of row into rw. Instead it treats it as a closure over the variable row itself:

for (var __index0__ = 0; __index0__ < __iter0__.length; __index0__++) {
    var r = __iter0__ [__index0__];
    var row = rows.row (r);
    deferreds.append ($.post ('assign_to_run', dict ({'run': run, 'sample_data': JSON.stringify (list ([row.data () [12]]))})).done ((function __lambda__ () {
        var rw = row;
        return rw.deselect ();})));
    }

When the above javascript code is run it will call deselect() on the last value assigned to row as many times as there were loop iterations, which isn't what it should be doing.

To make matters worse, Transcrypt moves evaluation of arguments into the function, which is wrong.

lambda *, rw=capture(row) : rw.deselect()

becomes when compiled:

function __lambda__ () {
    var rw = capture (row);
    return rw.deselect ();})));
}

proper behavior is shown by this code

def capture(v):
   val = v
   return v

fs = []
for i in [1,2,3]:
   fs.append(lambda *, ii=capture(i) : print(ii, i))

for f in fs:
   f()

which will print

1 3
2 3
3 3

Thanks for the thorough testing and analysis!
I'll try to fix this in the next version.

Thank you for taking such quick action on this, and thanks also for Transcrypt. It's very nice to be able to write client-side code in Python instead of ES5.

Well, to be honest, currently I am quite immersed in the source maps, so I'm afraid it will still be some time before the next version rolls out. I will try to do all the bug fixes in one go. Don't know if that succeeds.

As a first step I've tried to construct a minimal example to experiment with this:

results = []

for i in range (10):
    # results.append (lambda: i)
    results.append (lambda j = i: j)

print (results [0] ())

which generates:

var results = list ([]);
for (var i = 0; i < 10; i++) {
    results.append ((function __lambda__ (j) {
        if (typeof j == 'undefined' || (j != null && j .__class__ == __kwargdict__)) {;
            var j = i;
        };
        return j;}));
}
print (results [0] ());

Unfortunately I currently don't know a good and compact mechanism to make this work properly.
A workaround is to create a separate inner scope for your lambda:

results = []

for i in range (10):
    # results.append (lambda: i)                          # Will never work
    # results.append (lambda j = i: j)                   # Works in Python
    results.append ((lambda j: lambda: j) (i))      # Works in Transcrypt and Python

print (results [3] ())

I'll add this to the docs. For now consider the necessity of this workaround a limitation of Transcrypt. I've added the workaround as a testcase to the autotest including the comments, to document it for future moments of brightness...

Navigation Menu

Repository navigation

Bug in compiling lambda default argument values #36

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Bug in compiling lambda default argument values #36

Description

Activity

elgow commented on Apr 15, 2016

JdeH commented on Apr 15, 2016

elgow commented on Apr 15, 2016

JdeH commented on Apr 15, 2016

JdeH commented on Apr 21, 2016

JdeH commented on Apr 21, 2016

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions