Cannot insert blank line in global insert mode

[mohd-akram]

To reproduce:

$ cat file
enum
$ printf '%s\n' 'g/enum/i\' '\' '.' '%p' | nvi -e file

Expected output:

enum

Actual output:

enum

The correct behavior is observed in ed and classic vi/ex (tested on OpenIndiana 2024.04).

[lichray]

The debugger leads me to

nvi2/ex/ex_append.c

Lines 110 to 147 in 52c07e8

	/*
	* !!!
	* Anything that was left after the command separator becomes part
	* of the inserted text. Apparently, it was common usage to enter:
	*
	* :g/pattern/append|stuff1
	*
	* and append the line of text "stuff1" to the lines containing the
	* pattern. It was also historically legal to enter:
	*
	* :append|stuff1
	* stuff2
	* .
	*
	* and the text on the ex command line would be appended as well as
	* the text inserted after it. There was an historic bug however,
	* that the user had to enter *two* terminating lines (the '.' lines)
	* to terminate text input mode, in this case. This whole thing
	* could be taken too far, however. Entering:
	*
	* :append|stuff1\
	* stuff2
	* stuff3
	* .
	*
	* i.e. mixing and matching the forms confused the historic vi, and,
	* not only did it take two terminating lines to terminate text input
	* mode, but the trailing backslashes were retained on the input. We
	* match historic practice except that we discard the backslashes.
	*
	* Input lines specified on the ex command line lines are separated by
	* <newline>s. If there is a trailing delimiter an empty line was
	* inserted. There may also be a leading delimiter, which is ignored
	* unless it's also a trailing delimiter. It is possible to encounter
	* a termination line, i.e. a single '.', in a global command, but not
	* necessary if the text insert command was the last of the global
	* commands.
	*/

Is this document relevant?

[mohd-akram]

Yes, somewhat, I think this is just a bug. A possible fix is to just remove the if (t != p || len == 0) check (I suggested it to the nvi maintainers to validate it), but it seems correct.