128059b92 ("hush: fix segfault on syntax error") attempted to fix a
segfault on syntax errors, but it broke Ctrl-C handling, and the
assumption that it made, that rcode could not be -1, is incorrect.
Revert this change.

Reported-by: Stephen Warren <swarren@wwwdotorg.org>
Reported-by: Przemyslaw Marczak <p.marczak@samsung.com>
Signed-off-by: Rabin Vincent <rabin@rab.in>

run_command() returns success even if the command had a syntax error;
correct this behaviour.

Signed-off-by: Rabin Vincent <rabin@rab.in>
Acked-by: Simon Glass <sjg@chromium.org)

Hush segfaults if it sees a syntax error while attempting to parse a
command:

 $ ./u-boot -c "'"
 ...
 syntax error
 Segmentation fault (core dumped)

This is due to a NULL pointer dereference of in_str->p in static_peek().
The problem is that the exit condition for the loop in
parse_stream_outer() checks for rcode not being -1, but rcode is only
ever 0 or 1.

Signed-off-by: Rabin Vincent <rabin@rab.in>
Acked-by: Simon Glass <sjg@chromium.org)
Tested-by: Simon Glass <sjg@chromium.org)

Attempting to run:
 - an empty string
 - a string with just spaces

returns different error codes, 1 for the empty string and 0
for the string with just spaces.  Make both of them return
0 for consistency.

Signed-off-by: Rabin Vincent <rabin@rab.in>
Acked-by: Simon Glass <sjg@chromium.org)

The run command treats each argument an an environment variable. It gets the
value of each variable and executes it as a command. If an environment
variable contains a newline and the hush cli is used, it is supposed to
execute each line one after the other.

Normally a newline signals to hush to exit - this is used in normal command
line entry - after a command is entered we want to return to allow the user
to enter the next one. But environment variables obviously need to execute
to completion.

Add a special case for the execution of environment variables which
continues when a newline is seen, and add a few tests to check this
behaviour.

Note: it's not impossible that this may cause regressions in other areas.
I can't think of a case but with any change of behaviour with limited test
coverage there is always a risk. From what I can tell this behaviour has
been around since at least U-Boot 2011.03, although this pre-dates sandbox
and I have not tested it on real hardware.

Reported-by: Wolfgang Denk <wd@denx.de>
Signed-off-by: Simon Glass <sjg@chromium.org>

Clang interpretes an if condition like  "if ((a = b) == NULL)
as it tries to assign a value in a statement. Hence if you do
"if ((something)) it warns you that you might be confused.
Hence drop the double braces for plane if statements.

Simon Glass <sjg@chromium.org>
Signed-off-by: Jeroen Hofstee <jeroen@myspectrum.nl>

When a simple command like 'false' is provided, hush should return the
result of that command. However, hush only does this if the
FLAG_EXIT_FROM_LOOP flag is provided. Without this flag, hush will
happily execute the empty string command immediate after 'false' and
then return a success code.

This behaviour does not seem very useful, and requiring the flag also
seems wrong, since it means that hush will execute only the first command
in a sequence.

Add a check for empty string and fall out of the loop in that case. That
at least fixes the simple command case. This is a change in behaviour but
it is unlikely that the old behaviour would be considered correct in any
case.

Reported-by: Stefan Herbrechtsmeier <stefan@herbrechtsmeier.net>
Signed-off-by: Simon Glass <sjg@chromium.org>

Add a bootretry_ prefix to these two functions, and remove the need for
the #ifdef around everything (it moves to the Makefile).

Signed-off-by: Simon Glass <sjg@chromium.org>

This code is only used by one board, so it seems a shame to clutter up
the readline code with it. Move it into its own file.

Signed-off-by: Simon Glass <sjg@chromium.org>

This makes it clear where the code resides.

Signed-off-by: Simon Glass <sjg@chromium.org>