[shell] improve mawk detection

[step-]

• Use the all-compatible mawk -W version option. replace awk for perl in the bash history widget #3313 (comment).
• Do not remap the history key if no dependent commands is installed (perl, awk or mawk in this order).
• Run the command and not a function consistently with use command to “protect” further commands #3462.

The version check bash code relies on the following mawk source code, extracted from mawk 1.3.4 20230322.

version.c:
18-  #include "init.h"
19-  #include "patchlev.h"
20-
21:  #define	 VERSION_STRING	 \
22-    "mawk %d.%d%s %s\n\
23-  Copyright 2008-2022,2023, Thomas E. Dickey\n\
24-  Copyright 1991-1996,2014, Michael D. Brennan\n\n"
....
30-  void
31-  print_version(FILE *fp)
32-  {
33:      fprintf(fp, VERSION_STRING, PATCH_BASE, PATCH_LEVEL, PATCH_STRING, DATE_STRING);
34-      fflush(fp);
35-
36-  #define SHOW_RANDOM "random-funcs:"

patchlev.h:
13-  /*
14-   * $MawkId: patchlev.h,v 1.128 2023/03/23 00:23:57 tom Exp $
15-   */
16:  #define  PATCH_BASE	1
17-  #define  PATCH_LEVEL	3
18-  #define  PATCH_STRING	".4"
19-  #define  DATE_STRING    "20230322"

[calestyo]

Other than the comments made in the code above, I think it looks mostly good, I especially like that you saved at least one execution of awk, though:

• I'd split it up in more commits, you do now various mostly unrelated things in one commit, adding the command, fixing the version check, changing the output redirection (which I would however drop completely), make the code handle the situation when both perl and [m]awk aren't available.
  I think it's easier to understand and follow in smaller pieces
• Handling the case that awk isn't available is in principle fine by me, though at least POSIX wise, systems are expected to have awk. So assuming POSIX systems, handling this is unnecessary.
• btw: I think the (x *1000 +y) *1000 +z >= 1003004 part is at least a bit fragile, should e.g. ever a big minor version number be used. Also you still have the check for the date, so apart from me not liking such a runtime check in general, t should be fine.

I checked it on Debian with mawk 1.3.4.20200120-3.1, which yields __fzf_awk=awk, and with 1.3.4.20230808-1, which yields __fzf_awk=mawk. So that works :-)

[step-]

@calestyo thank you for testing on Debian.
@junegunn you're right, who can be sure how all target platforms and bash versions will behave when stdout is closed. Better be safe.

[junegunn]

Trying to remove unnecessary diffs.

[junegunn]

• Handling the case that awk isn't available is in principle fine by me, though at least POSIX wise, systems are expected to have awk. So assuming POSIX systems, handling this is unnecessary.

I also think it's unnecessary. The absence of awk can be a sign of a broken system, and we should probably warn the user with an error instead of silently disabling a subset of our feature, making it harder to fix the system.

[junegunn]

Merged, thanks!

[calestyo]

Looks good to me, though I probably still have placed the command inside the variable, as described above.

[calestyo]

Oh, one more thing that I had noticed before, but forgot to write down:

fzf/shell/key-bindings.bash

Line 75 in 8848560

IFS=' .' read n x y z d <<< $(command mawk -W version 2> /dev/null)

Wouldn't it be safer to use read -r?