(Travis CI timed out.)

Just to confirm: modified was only set on nodes in figure_out_scope previously to be used solely by the reduce_vars optimization? So removing modified from figure_out_scope won't negatively affect any code?

@kzc yup - I introduced it back in #1301, and I double-checked and made sure nobody else has decided to utilise this yet.

I'm positively surprised that reduce_vars handles parameters to IIFEs. Can you confirm this parameter optimization is limited to IIFEs and not functions in general?

This was successfully reduced:

$ echo '!function(a){console.log(a-1);}(42);' | bin/uglifyjs -c reduce_vars
!function(a){console.log(41)}(42);

but oddly this was not:

$ echo '!function(a){console.log(a);}(42);' | bin/uglifyjs -c reduce_vars
!function(a){console.log(a)}(42);

Edit: was it because the result is shorter?

nope:

$ echo '!function(ARGUMENT){console.log(ARGUMENT);}(42);' | bin/uglifyjs -c reduce_vars
!function(ARGUMENT){console.log(ARGUMENT)}(42);

but oddly this was not:

That is a known quirk with evaluate:

$ echo 'const a=1;console.log(a+1)' | bin/uglifyjs -c
const a=1;console.log(2);

$ echo 'const a=1;console.log(a)' | bin/uglifyjs -c
const a=1;console.log(a);

Can you confirm this parameter optimization is limited to IIFEs and not functions in general?

I borrow the logic from negate_iife, so am fairly certain this covers only IIFEs - the normal form, and one which does new function(...){...}(...)

Another two quirks I've discovered is to do with keep_fnames and keep_fargs, both of which I have an idea to improve upon in an upcoming PR 😉

Edit: #1476

Discovered by @kzc while testing with #1479, the following won't work with reduce_vars on master (& probabaly 2.7.5 as well):

if (code == 16) {
  var bitsLength = 2, bitsOffset = 3, what = len;
} else if (code == 17) {
  var bitsLength = 3, bitsOffset = 3, what = (len = 0);
} else if (code == 18) {
  var bitsLength = 7, bitsOffset = 11, what = (len = 0);
}
var repeatLength = this.getBits(bitsLength) + bitsOffset;

Note the repeated var definitions of bitsOffset. I'm thinking of including the fix in this PR to reduce potential merge conflicts.

uglify-js@2.7.5 produces the correct last line with -c reduce_vars:

var repeatLength = this.getBits(bitsLength) + bitsOffset;

master 7f8d72d produces:

var repeatLength = this.getBits(bitsLength) + 11;

reduce_vars regression introduced in 0610c02

Specifically, these lines.

But my new test case in f42a5db will cause even 2.7.5 to fail 😜

On the positive side, your new test/jetstream.js script is doing what it was designed to do.

Can you squash + change commit description to "improve reduce_vars and fix a bug" ?

@alexlamsl Would you mind adding the test case from #1473 (comment) since it apparently exercises different logic than the test case you committed?

@kzc test added to squashed commit

(Travis CI, please try again? Thank you.)