``make check-install'' verses ``make check''

[Loren James Rittle <rittle at latour dot rsch dot comm dot mot dot com>]

Ignoring all size and time differences, I have noticed the following
difference between ``make check'' and ``make check-install'' output on
i386-unknown-freebsd4.2 (the only operation run on the source or
object tree between checks was ``make install''):

< pass/fail results:  102/4 shared + 101/5 static = 203/9 total
---
> pass/fail results:  102/4 shared + 100/6 static = 202/10 total

< +  9  0.000   162400  7432    2086716 21_strings/insert.cc -static
< +  7  0.000   25896   1820    269405  21_strings/insert.cc 
---
> -r 10 0.000   154568  7400    2218219 21_strings/insert.cc -static
> +  7  0.000   14496   88      186120  21_strings/insert.cc 

[BTW, 21_strings/ctor_copy_dtor.cc is the extra static failure.  It
 has appeared and disappeared over the past week/month but I have been
 unable to correlate its appearance to any particular configuration
 change.  --with-dwarf2 is the only configuration option I have been
 adding and removing but those tests all force -ggdb3 anyways...]

And this difference on alpha-unknown-freebsd4.2 (annoyingly enough, it
is a "different" difference than above):

< pass/fail results:  95/11 shared + 102/4 static = 197/15 total
---
> pass/fail results:  94/12 shared + 101/5 static = 195/17 total

< +  12 0.000   392048  3072    2625100 22_locale/ctype_char_members.cc -static
< +  11 0.000   27328   2872    259323  22_locale/ctype_char_members.cc 
---
> -r 11 0.000   378240  3040    2849894 22_locale/ctype_char_members.cc -static
> -r 10 0.000   7632    8       135546  22_locale/ctype_char_members.cc 

[BTW, the extra 7 shared failures on this platform are all related to
 a simple g++ test case involving exception handling that doesn't
 appear to work on alpha-unknown-freebsd4.2.]

I immediately noticed one oddity while debugging these inconsistencies:

$ ldd testsuite/insert.st-exe
testsuite/insert.st-exe:
        libm.so.2 => /usr/lib/libm.so.2 (0x280af000)
        libstdc++.so.3 => /usr/local/beta-gcc/lib/libstdc++.so.3 (0x280cb000)
        libc.so.4 => /usr/lib/libc.so.4 (0x28154000)

One could ask: Why is the so-called "static build" still linking in
the shared version of the library as well?  However, editing mkcheck
to force -all-static instead of -static revealed that this was not the
root cause of the problem (at least it still crashed with the same
backtrace as before the change).

Is anyone else seeing this problem on any other platforms (do others
even bother running check-install ;-)?

Regards,
Loren