Improve indentation: bugfix, heredoc, embdoc, strings

[tompng]

Description

• Implement heredoc and embdoc indent
• Implement string indent
• Fixes all indentation bug that I know now

left: irb 1.6.2 right: this branch

perfect_indent.mp4

Heredoc and embdoc indent

I implemented these heredoc and embdoc indent

if 1 x = <<EOS heredoc # initial indent should be 0 EOS puts x =begin embdoc # initial indent should be 0 =end y = <<~EOS1 + <<~EOS2  heredoc1 # initial indent should be 4  # indent of heredoc_end should be 2  EOS1  heredoc2  EOS2 end

<<~JS  function f() {  const x = 1 // any number of indent(>=2) should be accepted  console.log(x) // next-line's initial indent should be same as prev line  } JS

String indent

string, backtick, regexp and symbol

def send_response # Inside string literal, any number of indent(>=0) should be accepted. # In master branch, two spaces will automatically inserted and make http header invalid. http_header = "HTTP/1.1 200 OK Content-Type: text/html Content-Length: #{body.bytesize}" socket.write "#{http_header}\n\n#{body}" end

Indentation bugfix

These bugs will be fixed

> if 1 > def f() = (█1 * 2) > 1 # ↓ insert line-break by pressing option+enter > if 1 > def f() = ( > █1 * 2) > 1

> if 1 > a = 1█ > 1 # ↓ insert line-break by pressing option+enter > if 1 > a = 1 > █ > 1

if 1 beginning en█ing # ↓ type d if 1 beginning end█ing

Internal

auto_indent_proc ->(lines, line_index, byte_pointer, is_newline){} was implemented as:

if is_newline # next line's indent number else # current line's indent if indent number is decreasing # nil (means any number of indent is accepted) if indent number is not decreasing (this will cause a bug) # note that it ignores right side of the cursor in the line which also cause a bug end

But it is wrong. In most of the cases, it should return a number to avoid bug when inserting newline.

if is_newline # next line's indent number else # nil (means any number of indent is accepted) when deleting or adding indent spaces # checked by left side of the cursor is all space or not # otherwise, returns current line's indent end

Also, special treatment is done for heredoc, embdoc and free-indent-acceptable literal like string.

[tompng]

This indent retain feature dropped in this pull request had a bug.

Paste this code and press return key

a .b .c if d

then you will get wrong indent

irb(main):001:0> a irb(main):002:0> .b irb(main):003:0> .c irb(main):004:1* if d irb(main):005:1* 

[st0012]

Does this mean lines is either [nil] or [""] when the condition is true?

[tompng]

Nice catch, it looks like almost-unreachable code.

I think I added this to returns nil if lines are empty, but empty lines are not passed to this proc

[tompng]

When I type CTRL+d, [nil] is passed. It might be a bug of reline.
I added again with more simple condition and comment.

next nil if lines == [nil] # Workaround for exit IRB with CTRL+d

[tompng]

I changed this condition simpler to if prev_opens.size <= next_opens.size

When heredoc is closed, next_opens.size is always smaller than prev_opens.size because only heredoc_end can be on the line.

<<A # heredoc_beg A + ( # heredoc_end + lparen ← this is not possible A ↑ heredoc ends here