The fill commands can deduce the proper fill prefix for a paragraph automatically in certain cases: either whitespace or certain punctuation characters at the beginning of a line are propagated to all lines of the paragraph.
If the paragraph has two or more lines, the fill prefix is taken from the paragraph’s second line, but only if it appears on the first line as well.
If a paragraph has just one line, fill commands may take a prefix from that line. The decision is complicated because there are three reasonable things to do in such a case:
All three of these styles of formatting are commonly used. So the fill commands try to determine what you would like, based on the prefix that appears and on the major mode. Here is how.
If the prefix found on the first line matches adaptive-fill-first-line-regexp, or if it appears to be a comment-starting sequence (this depends on the major mode), then the prefix found is used for filling the paragraph, provided it would not act as a paragraph starter on subsequent lines.
Otherwise, the prefix found is converted to an equivalent number of spaces, and those spaces are used as the fill prefix for the rest of the lines, provided they would not act as a paragraph starter on subsequent lines.
In Text mode, and other modes where only blank lines and page delimiters separate paragraphs, the prefix chosen by adaptive filling never acts as a paragraph starter, so it can always be used for filling.
The variable adaptive-fill-regexp determines what kinds of line
beginnings can serve as a fill prefix: any characters at the start of
the line that match this regular expression are used. If you set the
variable adaptive-fill-mode to
nil, the fill prefix is
never chosen automatically.
You can specify more complex ways of choosing a fill prefix
automatically by setting the variable adaptive-fill-function to a
function. This function is called with point after the left margin of a
line, and it should return the appropriate fill prefix based on that
line. If it returns
nil, adaptive-fill-regexp gets
a chance to find a prefix.