Machine Type: x86_64-unknown-linux-gnu
Bash Version: 4.4
Patch Level: 12
Release Status: release
Description:
When a function looks like it is defined in multiple sources,
BASH_SOURCE does not reliably report it has defined
where it was actually defined. It seems to prefer the
last source bash has visited.
This is not related to a multiple definition. When I
say 'looks like', it actually 'looks like' from an
external point of view, but it does not from a
runtime point of view.
Running the attached example, we get:
$ ./bash-4.4.12 call-stack.sh
1 def-2.sh:5 def-1.sh x_trace_stack
1 def-2.sh:5 def-2.sh x_trace_stack
The cells 'def-2.sh:5' are build from BASH_SOURCE and
BASH_LINENO. They should report in order:
def-1.sh:5
def-2.sh:5
but the first one is wrong. Note that BASH_LINENO is correct,
and this is unlikely due to chance.
Repeat-By:
Here are three scripts:
- call-stack.sh
- def-1.sh
- def-2.sh
Both def-1.sh and def-2.sh /may/ define functions 'def1'
and 'def2',
but actually def-1.sh defines 'def1' while def-2.sh
defines 'def2',
as those functions do report themselves.
Sourcing call-stack.sh, we observe that both 'def1' and
'def2' are reported by BASH_SOURCE to be defined in def-2.sh.
Scripts:
# ---------------------------------------------------------
# call-stack.sh
#!/bin/bash
x_trace_stack() { # [MESSAGE START STOP] # Dump call stack #1
# START STOP unused
local -i i=1
local ONCE="$1"; shift
printf "%d %15s:%-3s %8s %s\n" \
$i \
"${BASH_SOURCE[i]}" \
"${BASH_LINENO[i-1]}" \
"$ONCE" \
"${FUNCNAME[i-1]}"
}
. def-1.sh
. def-2.sh
def1
def2
# ------------------------------------------------------------
# def-1.sh
#!/bin/bash
_() { [ $1 = def1 ]; }
_ def1 && def1() { x_trace_stack def-1.sh 1 2; }
_ def2 && def2() { x_trace_stack def-1.sh 1 2; }
# ------------------------------------------------------------
# def-2.sh
#!/bin/bash
_() { [ $1 = def2 ]; }
_ def2 && def2() { x_trace_stack def-2.sh 1 2; }
_ def1 && def1() { x_trace_stack def-2.sh 1 2; }
[Prev in Thread]
Current Thread
[Next in Thread]
BASH_SOURCE gets confused when multiple sources are a priori candidates,
Bruno Vasselle<=