This is a question rather than a bug report. Four tests of the regression
suite fail here on a clean pre-master-47, before any local change, and we
would like to know whether they pass for you, so that we can tell whether
our environment is at fault.
cd build && make -k check
FAIL: ex1a.cir
FAIL: ex1b.cir
FAIL: ex2a.cir
FAIL: ex3a.cir
all four in tests/regression/lib-processing, everything else passes (65
tests pass with our patch series applied, 54 on the bare tree).
The output of ex1a.cir:
File included as: .inc 'ex1.lib' RES
File included as: .inc RES
Error: Could not find include file RES
While reading ../../../../tests/regression/lib-processing/./ex1.lib
ERROR: fatal error in ngspice, exit(1)
so the section name of the .lib line, RES, is taken as the name of the
file to include. ex2a and ex3a fail the same way with the section MOS.
It is not a vpath artifact of our out of tree build: running the deck from
inside the source directory, with the source directory as the working
directory, fails identically.
cd tests/regression/lib-processing
../../../build/src/ngspice --batch ex1a.cir
pre-master-47?If they do, is there a configure option or an environment variable which
affects the handling of .lib with a section name, and which we are
missing? We configured with
../configure --with-x --enable-xspice --enable-cider --enable-openmp \
--with-readline=no CFLAGS="-O1 -g -Wall"
We did not investigate further, because the failures are unrelated to the
work we are submitting, and we did not want to change behaviour we do not
understand. All results reported in the accompanying series are stated
against this baseline: the same four tests fail before and after our
patches, and no other test changes state.
ngspice pre-master-47, x86_64 Linux (WSL2, kernel 6.6), gcc 13.3,
bison 3.8.2, flex 2.6.4, automake 1.16.5, autoconf 2.71
Running make check requires a
set ngbehavior=mcin .spiceinit (or spice.rc)