Thread: Problems with src/pl/tcl/mkMakefile.tcldefs.sh.in in 6.5
Problems with src/pl/tcl/mkMakefile.tcldefs.sh.in in 6.5
From
pgsql-hackers@thewrittenword.com
Date:
For Digital UNIX 4.0D, shared libraries are created by:$ ld -shared -expect_unresolved "*" -o foo.so [objects] This presents a problem for mkMakefile.tcldefs.sh.in. In tclConfig.sh:TCL_SHLIB_LD='ld -shared -expect_unresolved "*"' In mkMakefile.tcldefs.sh.in:cat @TCL_CONFIG_SH@ |egrep '^TCL_|^TK_' |while read inpdo eval eval echo $inpdone >Makefile.tcldefs Because of this, we wind up with the following in Makefile.tcldefs to created shared libraries on Digital UNIX because of the eval:TCL_SHLIB_LD=ld -shared -expect_unresolved * The "*" needs to be quoted to avoid shell expansion. How about the following:cat @TCL_CONFIG_SH@ |egrep '^TCL_|^TK_' |sed -e "s/^\([^=]*\)='\(.*\)'$/\1=\2/" -- albert chin (china@thewrittenword.com)
I didn't understand this the first time you sent it either. Send me a patch to review, please. > For Digital UNIX 4.0D, shared libraries are created by: > $ ld -shared -expect_unresolved "*" -o foo.so [objects] > > This presents a problem for mkMakefile.tcldefs.sh.in. In tclConfig.sh: > TCL_SHLIB_LD='ld -shared -expect_unresolved "*"' > > In mkMakefile.tcldefs.sh.in: > cat @TCL_CONFIG_SH@ | > egrep '^TCL_|^TK_' | > while read inp > do > eval eval echo $inp > done >Makefile.tcldefs > > Because of this, we wind up with the following in Makefile.tcldefs to > created shared libraries on Digital UNIX because of the eval: > TCL_SHLIB_LD=ld -shared -expect_unresolved * > > The "*" needs to be quoted to avoid shell expansion. How about the > following: > cat @TCL_CONFIG_SH@ | > egrep '^TCL_|^TK_' | > sed -e "s/^\([^=]*\)='\(.*\)'$/\1=\2/" > > -- > albert chin (china@thewrittenword.com) > > -- Bruce Momjian | http://www.op.net/~candle maillist@candle.pha.pa.us | (610) 853-3000+ If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania19026
Re: [HACKERS] Problems with src/pl/tcl/mkMakefile.tcldefs.sh.in in 6.5
From
pgsql-hackers@thewrittenword.com
Date:
On Sat, Jul 10, 1999 at 02:45:47AM -0400, Bruce Momjian wrote: > I didn't understand this the first time you sent it either. > > Send me a patch to review, please. --- src/pl/tcl/mkMakefile.tcldefs.sh.in.orig Fri Jul 9 08:29:09 1999 +++ src/pl/tcl/mkMakefile.tcldefs.sh.in Fri Jul 9 08:29:49 1999 @@ -8,9 +8,6 @@cat @TCL_CONFIG_SH@ | egrep '^TCL_|^TK_' | - while read inp - do - eval eval echo $inp - done >Makefile.tcldefs + sed -e "s/^\([^=]*\)='\(.*\)'$/\1=\2/" >Makefile.tcldefsexit 0 > > > For Digital UNIX 4.0D, shared libraries are created by: > > $ ld -shared -expect_unresolved "*" -o foo.so [objects] > > > > This presents a problem for mkMakefile.tcldefs.sh.in. In tclConfig.sh: > > TCL_SHLIB_LD='ld -shared -expect_unresolved "*"' > > > > In mkMakefile.tcldefs.sh.in: > > cat @TCL_CONFIG_SH@ | > > egrep '^TCL_|^TK_' | > > while read inp > > do > > eval eval echo $inp > > done >Makefile.tcldefs > > > > Because of this, we wind up with the following in Makefile.tcldefs to > > created shared libraries on Digital UNIX because of the eval: > > TCL_SHLIB_LD=ld -shared -expect_unresolved * > > > > The "*" needs to be quoted to avoid shell expansion. How about the > > following: > > cat @TCL_CONFIG_SH@ | > > egrep '^TCL_|^TK_' | > > sed -e "s/^\([^=]*\)='\(.*\)'$/\1=\2/" > > > > -- > > albert chin (china@thewrittenword.com) > > -- > Bruce Momjian | http://www.op.net/~candle -- albert chin (china@thewrittenword.com)
> On Sat, Jul 10, 1999 at 02:45:47AM -0400, Bruce Momjian wrote: > > I didn't understand this the first time you sent it either. > > > > Send me a patch to review, please. > > --- src/pl/tcl/mkMakefile.tcldefs.sh.in.orig Fri Jul 9 08:29:09 1999 > +++ src/pl/tcl/mkMakefile.tcldefs.sh.in Fri Jul 9 08:29:49 1999 > @@ -8,9 +8,6 @@ > > cat @TCL_CONFIG_SH@ | > egrep '^TCL_|^TK_' | > - while read inp > - do > - eval eval echo $inp > - done >Makefile.tcldefs > + sed -e "s/^\([^=]*\)='\(.*\)'$/\1=\2/" >Makefile.tcldefs > I understand what your patch does, and it looks OK, but any idea why the 'eval eval' was there, and is it safe to skip it? I can apply this to 6.6. -- Bruce Momjian | http://www.op.net/~candle maillist@candle.pha.pa.us | (610) 853-3000+ If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania19026
Bruce Momjian <maillist@candle.pha.pa.us> writes:
> I understand what your patch does, and it looks OK, but any idea why the
> 'eval eval' was there, and is it safe to skip it?
I think the idea is to expand any shell variable references that appear
in tclConfig.sh. If so, the code is wrong anyway, since the expansion
will occur in a subshell that hasn't actually executed tclConfig.sh,
and therefore does not have definitions for the referenced variables.
We'll always get empty strings substituted for the references, and that
may be the wrong thing.
Taking an example that actually appears in my system's tclConfig.sh:
TCL_LIB_FILE='libtcl8.0${TCL_DBGX}.a'
I stick this into shell variable inp:
$ read inp
TCL_LIB_FILE='libtcl8.0${TCL_DBGX}.a' --- typed by me
$ echo $inp
TCL_LIB_FILE='libtcl8.0${TCL_DBGX}.a'
Now one eval gets rid of the outer single quotes:
$ eval echo $inp
TCL_LIB_FILE=libtcl8.0${TCL_DBGX}.a
and another one will perform a round of shell interpretation on what's
inside the quotes:
$ eval eval echo $inp
TCL_LIB_FILE=libtcl8.0.a
which is not what we want. In this particular case it's harmless
because TCL_DBGX should be empty, but if I had a debugging build
of Tcl installed here then the makefile would fail because it would
have the wrong value of TCL_LIB_FILE.
If we do something like what Albert is proposing, the sed script
will need to convert ${...} to $(...) so that shell variable references
become make variable references.
regards, tom lane
Re: [HACKERS] Problems with src/pl/tcl/mkMakefile.tcldefs.sh.in in 6.5
From
pgsql-hackers@thewrittenword.com
Date:
On Sat, Jul 10, 1999 at 02:45:47AM -0400, Bruce Momjian wrote: > I didn't understand this the first time you sent it either. > > Send me a patch to review, please. --- src/pl/tcl/mkMakefile.tcldefs.sh.in.orig Fri Jul 9 08:29:09 1999 +++ src/pl/tcl/mkMakefile.tcldefs.sh.in Fri Jul 9 08:29:49 1999 @@ -8,9 +8,6 @@cat @TCL_CONFIG_SH@ | egrep '^TCL_|^TK_' | - while read inp - do - eval eval echo $inp - done >Makefile.tcldefs + sed -e "s/^\([^=]*\)='\(.*\)'$/\1=\2/" >Makefile.tcldefsexit 0 > > > For Digital UNIX 4.0D, shared libraries are created by: > > $ ld -shared -expect_unresolved "*" -o foo.so [objects] > > > > This presents a problem for mkMakefile.tcldefs.sh.in. In tclConfig.sh: > > TCL_SHLIB_LD='ld -shared -expect_unresolved "*"' > > > > In mkMakefile.tcldefs.sh.in: > > cat @TCL_CONFIG_SH@ | > > egrep '^TCL_|^TK_' | > > while read inp > > do > > eval eval echo $inp > > done >Makefile.tcldefs > > > > Because of this, we wind up with the following in Makefile.tcldefs to > > created shared libraries on Digital UNIX because of the eval: > > TCL_SHLIB_LD=ld -shared -expect_unresolved * > > > > The "*" needs to be quoted to avoid shell expansion. How about the > > following: > > cat @TCL_CONFIG_SH@ | > > egrep '^TCL_|^TK_' | > > sed -e "s/^\([^=]*\)='\(.*\)'$/\1=\2/" > > > > -- > > albert chin (china@thewrittenword.com) > > -- > Bruce Momjian | http://www.op.net/~candle -- albert chin (china@thewrittenword.com)
Re: [ADMIN] Re: [HACKERS] Problems with src/pl/tcl/mkMakefile.tcldefs.sh.in in 6.5
From
wieck@debis.com (Jan Wieck)
Date:
Bruce Momjian wrote:
>
> > On Sat, Jul 10, 1999 at 02:45:47AM -0400, Bruce Momjian wrote:
> > > I didn't understand this the first time you sent it either.
> > >
> > > Send me a patch to review, please.
> >
> > --- src/pl/tcl/mkMakefile.tcldefs.sh.in.orig Fri Jul 9 08:29:09 1999
> > +++ src/pl/tcl/mkMakefile.tcldefs.sh.in Fri Jul 9 08:29:49 1999
> > @@ -8,9 +8,6 @@
> >
> > cat @TCL_CONFIG_SH@ |
> > egrep '^TCL_|^TK_' |
> > - while read inp
> > - do
> > - eval eval echo $inp
> > - done >Makefile.tcldefs
> > + sed -e "s/^\([^=]*\)='\(.*\)'$/\1=\2/" >Makefile.tcldefs
> >
>
> I understand what your patch does, and it looks OK, but any idea why the
> 'eval eval' was there, and is it safe to skip it? I can apply this to
> 6.6.
As far as I can recall, the first of all versions I've
created did it mainly that way (with a simple sed(1) call).
But since tclConfig.sh is a shell script, there have to be
shell variable expansions done on some platforms and that
resulted finally in the double eval. So I would consider the
above a little step for a man, but a big leap backward for
mankind.
Instead, the result of the double eval must get special
characters quoted in some way.
Jan
--
#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me. #
#========================================= wieck@debis.com (Jan Wieck) #
Re: [ADMIN] Re: [HACKERS] Problems with src/pl/tcl/mkMakefile.tcldefs.sh.in in 6.5
From
pgsql-hackers@thewrittenword.com
Date:
On Sun, Jul 11, 1999 at 03:46:26PM +0200, Jan Wieck wrote:
> Bruce Momjian wrote:
>
> >
> > > On Sat, Jul 10, 1999 at 02:45:47AM -0400, Bruce Momjian wrote:
> > > > I didn't understand this the first time you sent it either.
> > > >
> > > > Send me a patch to review, please.
> > >
> > > --- src/pl/tcl/mkMakefile.tcldefs.sh.in.orig Fri Jul 9 08:29:09 1999
> > > +++ src/pl/tcl/mkMakefile.tcldefs.sh.in Fri Jul 9 08:29:49 1999
> > > @@ -8,9 +8,6 @@
> > >
> > > cat @TCL_CONFIG_SH@ |
> > > egrep '^TCL_|^TK_' |
> > > - while read inp
> > > - do
> > > - eval eval echo $inp
> > > - done >Makefile.tcldefs
> > > + sed -e "s/^\([^=]*\)='\(.*\)'$/\1=\2/" >Makefile.tcldefs
> > >
> >
> > I understand what your patch does, and it looks OK, but any idea why the
> > 'eval eval' was there, and is it safe to skip it? I can apply this to
> > 6.6.
>
> As far as I can recall, the first of all versions I've
> created did it mainly that way (with a simple sed(1) call).
> But since tclConfig.sh is a shell script, there have to be
> shell variable expansions done on some platforms and that
> resulted finally in the double eval. So I would consider the
> above a little step for a man, but a big leap backward for
> mankind.
>
> Instead, the result of the double eval must get special
> characters quoted in some way.
I just looked at the man for make on Solaris, Digital UNIX, HP-UX, and
IRIX and all support $() and ${} for variable expansion. BTW, I also
looked at the Makefile generated by Tk and it assumes make can handle
${}. It basically does one eval of tclConfig.sh and uses the result in
make variables. As Tk assumes make can handle ${}, can we safely
assume the same? With this, we'd do one eval rather than two before.
This is the same as the sed line I posted because everything in
tclConfig.sh is VAR='VAL' so the one eval would just strip the single
quotes (which sed did). But, if tclConfig.sh changes to some use of
double quotes in the future, we won't break.
> Jan
--
albert chin (china@thewrittenword.com)
I believe this is fixed. > For Digital UNIX 4.0D, shared libraries are created by: > $ ld -shared -expect_unresolved "*" -o foo.so [objects] > > This presents a problem for mkMakefile.tcldefs.sh.in. In tclConfig.sh: > TCL_SHLIB_LD='ld -shared -expect_unresolved "*"' > > In mkMakefile.tcldefs.sh.in: > cat @TCL_CONFIG_SH@ | > egrep '^TCL_|^TK_' | > while read inp > do > eval eval echo $inp > done >Makefile.tcldefs > > Because of this, we wind up with the following in Makefile.tcldefs to > created shared libraries on Digital UNIX because of the eval: > TCL_SHLIB_LD=ld -shared -expect_unresolved * > > The "*" needs to be quoted to avoid shell expansion. How about the > following: > cat @TCL_CONFIG_SH@ | > egrep '^TCL_|^TK_' | > sed -e "s/^\([^=]*\)='\(.*\)'$/\1=\2/" > > -- > albert chin (china@thewrittenword.com) > > -- Bruce Momjian | http://www.op.net/~candle maillist@candle.pha.pa.us | (610) 853-3000+ If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania19026
Re: [HACKERS] Problems with src/pl/tcl/mkMakefile.tcldefs.sh.in in 6.5
From
wieck@debis.com (Jan Wieck)
Date:
Bruce Momjian wrote:
>
> I believe this is fixed.
Again one of the frequently appearing items. So I would call
it more "hacked quiet - for now" instead of "fixed".
The script is only called if PostgreSQL is configured
--with_tcl. In that case, missing the Tcl(/Tk) includes
and/or libs would cause errors and a compilation abort. Can't
we assume that if the user configured with Tcl, she would at
least have a working tclsh(1)? I think we can. I don't know
of any "normal" Tcl-installation where the libs are present
but no working tclsh(1).
Since Tcl itself has much better capabilities than a sh(1) or
sed(1), it might be reasonable to source in the tclConfig.sh
into mkMakefile.tclsh.sh and pipe a "set" trough a tcl script
that does the real conversion into proper Makefile escaping.
An advantage would be that the Tcl script could check if the
version of the systems default tclsh(1) is the same as the
one in the choosen tclConfig.sh file and notice the user if
not. Using different Tcl versions in the libs and includes
than in the tclsh(1) executable could cause horrible
problems. I'm unhappy with the current libpgtcl for a long
time, but the changes I have in mind would make it
incompatible with pre-8.0 Tcl. So the changes will cause a
bunch of #if...#else...#endif that MUST match the later used
tclsh(1) at compile time or the dynamic loader of Tcl would
fail.
Jan
BTW: Is it only me or do others too wonder why their private
wish-list is sometimes longer than our official TODO?
>
> > For Digital UNIX 4.0D, shared libraries are created by:
> > $ ld -shared -expect_unresolved "*" -o foo.so [objects]
> >
> > This presents a problem for mkMakefile.tcldefs.sh.in. In tclConfig.sh:
> > TCL_SHLIB_LD='ld -shared -expect_unresolved "*"'
> >
> > In mkMakefile.tcldefs.sh.in:
> > cat @TCL_CONFIG_SH@ |
> > egrep '^TCL_|^TK_' |
> > while read inp
> > do
> > eval eval echo $inp
> > done >Makefile.tcldefs
> >
> > Because of this, we wind up with the following in Makefile.tcldefs to
> > created shared libraries on Digital UNIX because of the eval:
> > TCL_SHLIB_LD=ld -shared -expect_unresolved *
> >
> > The "*" needs to be quoted to avoid shell expansion. How about the
> > following:
> > cat @TCL_CONFIG_SH@ |
> > egrep '^TCL_|^TK_' |
> > sed -e "s/^\([^=]*\)='\(.*\)'$/\1=\2/"
> >
> > --
> > albert chin (china@thewrittenword.com)
--
#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me. #
#========================================= wieck@debis.com (Jan Wieck) #
> Jan > > BTW: Is it only me or do others too wonder why their private > wish-list is sometimes longer than our official TODO? That's bad. Our official TODO is very large. -- Bruce Momjian | http://www.op.net/~candle maillist@candle.pha.pa.us | (610) 853-3000+ If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania19026
wieck@debis.com (Jan Wieck) writes:
> Bruce Momjian wrote:
>> I believe this is fixed.
> Again one of the frequently appearing items. So I would call
> it more "hacked quiet - for now" instead of "fixed".
No, it's really fixed, or anyway Albert's complaint is fixed
(there was at least one other related complaint recently, too).
The problem was not whether we could find tclConfig.sh, it was
whether we were coping correctly with shell metacharacters in
the variable values. We were not, but I fixed that with a redesign
of the way the script read the file...
regards, tom lane
On Sat, Jul 10, 1999 at 11:37:24AM -0400, Tom Lane wrote:
> Bruce Momjian <maillist@candle.pha.pa.us> writes:
> > I understand what your patch does, and it looks OK, but any idea why the
> > 'eval eval' was there, and is it safe to skip it?
>
> I think the idea is to expand any shell variable references that appear
> in tclConfig.sh. If so, the code is wrong anyway, since the expansion
> will occur in a subshell that hasn't actually executed tclConfig.sh,
> and therefore does not have definitions for the referenced variables.
> We'll always get empty strings substituted for the references, and that
> may be the wrong thing.
>
> Taking an example that actually appears in my system's tclConfig.sh:
> TCL_LIB_FILE='libtcl8.0${TCL_DBGX}.a'
>
> I stick this into shell variable inp:
>
> $ read inp
> TCL_LIB_FILE='libtcl8.0${TCL_DBGX}.a' --- typed by me
> $ echo $inp
> TCL_LIB_FILE='libtcl8.0${TCL_DBGX}.a'
>
> Now one eval gets rid of the outer single quotes:
>
> $ eval echo $inp
> TCL_LIB_FILE=libtcl8.0${TCL_DBGX}.a
>
> and another one will perform a round of shell interpretation on what's
> inside the quotes:
>
> $ eval eval echo $inp
> TCL_LIB_FILE=libtcl8.0.a
>
> which is not what we want. In this particular case it's harmless
> because TCL_DBGX should be empty, but if I had a debugging build
> of Tcl installed here then the makefile would fail because it would
> have the wrong value of TCL_LIB_FILE.
>
> If we do something like what Albert is proposing, the sed script
> will need to convert ${...} to $(...) so that shell variable references
> become make variable references.
Ok, try the following patch:
--- src/pl/tcl/mkMakefile.tcldefs.sh.in.orig Sat Jul 10 14:09:52 1999
+++ src/pl/tcl/mkMakefile.tcldefs.sh.in Sat Jul 10 14:15:50 1999
@@ -6,11 +6,11 @@ exit 1fi
+# Strip outer quotes from variable and expand ${VAR} to $(VAR) for
+# interpretation by makecat @TCL_CONFIG_SH@ | egrep '^TCL_|^TK_' |
- while read inp
- do
- eval eval echo $inp
- done >Makefile.tcldefs
+ sed -e "s/^\([^=]*\)='\(.*\)'$/\1=\2/g" \
+ -e 's/\${\([^}][^}]*\)}/$(\1)/g' >Makefile.tcldefsexit 0
> regards, tom lane
Thanks Tom!
--
albert chin (china@thewrittenword.com)