[Geany-Devel] Some comments on b4n regex-indent

Colomban Wendling lists.ban at herbesfolles.org
Thu Jan 31 00:15:06 UTC 2013

Le 31/01/2013 00:28, Lex Trotman a écrit :
> On 31 January 2013 09:55, Colomban Wendling <lists.ban at herbesfolles.org> wrote:
>> Le 26/01/2013 03:49, Lex Trotman a écrit :
>>> [...]
>>>  Solutions left as an
>>> exercise for the reader since I don't have any ;)
>> No no, I told you *you* were supposed to give me the solutions ;)
> Well, as I said on IRC, I am thinking about it, but it is adding more regexen :)

If it's the answer…

>>> C language, which I assume represents all {} languages
>>> 1. Indents relative to the previous line, that means that (tabs, width=4):
>>> if(long &&
>>>    more){
>>>        blah;
>>> is wrong, blah is indented too far, and hard to fix because unindent
>>> remains offset (by the extra indent of "more") until the start of the
>>> line.  Maybe calculate new indent relative to the previous indent
>>> level, not the alignment?
>>> This is the same as current, and just as annoying :)
>> Yeah.  I think for this we need clever parenthesis matching.  I even
>> have some hard-coded logic locally, made a long time ago because I found
>> this annoying, but not yet polished... I probably could make it a plugin
>> one day -- and it'd give the "plugin for filtypes" idea to a test.
> Yeah, alignment needs to be considered (or ignored, but that means
> looking further back).

I don't think a regex will ever be able to get alignment correctly, so
we probably gotta need some code for this.

>>> 3. Typing } anywhere (outside comment) causes a fluctuating indent,
>>> annoying whilst editing.  Not really sure what is going on here, seems
>>> very dependent on whats on this and the last line, can't find a
>>> pattern.
>> I guess what you see is "normal".  "}" is declared as a "trigger
>> character" in the configuration, so chat it does it… triggering
>> autoindent for the current line.  And the algorithm is:
>> line indent = prev_line_indent + indent_after_prev_line -
>> unindent_after_prev_line + indent_current_line - unindent_current_line
>> so since the current line indentation is re-computed using the previous
>> line indentation as a basis and applying the rules on it, the
>> indentation may change even if it doesn't match (in which case the
>> indentation would be the same as the one on the prev line).
>> Maybe I should add an "only update if a rule for the current line
>> matched" check?  Not sure if it would go against something...
> Ah, yeah, one of the principles in my thinking is *stability*, ie you
> get the same answer so things don't move about like this.

The problem is that we can't really get both stability (ie not touching
stuff that don't match) *and* automagic undoing of improper indent if a
RE stopped matching (the famous "unindent me at 'end' while typing but
not at 'ending()'")...

>>> 4. probably part of the above:
>>> if(long &&
>>>    more){ do; it; }
>>> removes the alignment of "more" when } is typed.  But as I read the
>>> unindent regex, it shouldn't work unless the } is only preceded by
>>> whitespace?
>> As you guessed, same as 3.
> Which begs the question for 3 as well, if unindent regex is ^\\s*} how
> does that match? Where does it get the unindentation from?

There is no unindentation.  the indentation of line 2 is recomputed as
follows when you type the "}" (which simply triggers line autoindent):

indent of line 2 - 1: 0
indent after line 2 - 1: 0
unindent after line 2 - 1: 0
indent of line 2: 0
unindent of line 2: 0

so we do 0 + 0 - 0 + 0 - 0 = 0, so line 2 gets indentation of 0.  And
unfortunately it breaks your manual alignment.

>> All this said, I changed a 2 things lately:
>> 1) I made it so a newline only automatically triggers autoindent if the
>> line has the same indent as the previous one, so hopefully it won't lose
>> manual or brace matched indent.  Actually I'm not sure we should really
>> automagically reindent upon newline insertion; but it's convenient in
>> the way it allows not to have any character triggers, in which case the
>> indentation is performed on newline only and probably won't ever get
>> annoying (beside the fact it's annoying for the indent not to be fixed
>> on the fly).
> Well, don't we have to calculate the indent of the new line every
> time? (notice the space after new:) How will:
> if(a){
>     blah
> get the indent for blah?  Triggering on { won't help, "blah" hasn't
> been typed yet, in fact that line doesn't yet exist.

As I guessed on IRC before you even read the mail, I was unclear.  The
points I added here are about auto-indenting the *current* line, e.g.
the line to which the \n was appended;  it doesn't change the
computation of next line indentation (which obviously don't depend on

The idea is that you type (| being the cursor):

if (a) {\n|

here, the post-line code kicks in.  first, it checks whether it has to
re-compute the indentation of this line, and sees that no, it's the
first one so we can't do anything about it [1].  then, it checks what
indent to add to the newly created line, and finds a match on "{$", then
inserting one level of indentation:

if (a) {\n

you continue typing:

if (a) {\n

here, it sees that current line (2) has a different indent as previous
line (1) so decides not to try anything on the current line (that's the
new thing).  it also sees that the next line indent is the same as the
current one, because there is no match, so you get

if (a) {\n

Now you type a "}".

if (a) {\n

if the autoindent triggers include "}", autoindent of the current line
kicks in:  line 3-1 indent is 4, and there is a match for unindent
current line, so we remove 1 level: 4 - 4 = 0, and you get

if (a) {\n

however, if you *didn't* put "}" in the triggers, nothing happens and
you continue on typing:

if (a) {\n

since current line (3) has the same indent as the previous one (2), the
autoindent kicks in and finds a match for "unindent current line", that
it applies:

if (a) {\n

now the "indent next line" kicks in too, and since there is no match it
just gives the same indent (none):

if (a) {\n

>> 2) Since now \n don't really trigger auto-indent, I fixed handling of
>> \n and \r as character triggers, so one can configure the thing to force
>> reindentation upon newline.
> So might as well do it on newline anyway,

yes, that's the point, so one could write rules that fixes the indent on
newline no matter what, which can be good for e.g. Ruby and the "end"
thing, so an improper unindent (e.g. after ending()) could get fixed
back.  however, rules that may have false positive may not list this in
the triggers so a newline don't mass up with the user indentation.

> and also now windows line
> ends will do reindentation twice (\r and \n), sure hope your algorithm
> is stable :)

The algorithm is stable since it's based on the previous line
indentation and always apply the same rules; kicking it in twice with
the same line content will give the same result.


[1] writing this, I see that it's wrong: "indent this line" isn't took
into account here... but who cares, it's an example and I thing the
actual code even is cleverer than that.

More information about the Devel mailing list