23 Feb
2023
23 Feb
'23
2:16 a.m.
Good day all, figured I'd start a thread on this matter as I'm starting to piece
enough together to articulate the questions arising in my research.
So based on my analysis of the 3B20S UNIX 4.1 manual I've been working through, all
evidence points to the formalized SGS package and COFF originating tightly coupled to the
3B-20 line, then growing legs to support VAX, but never quite absorbing PDP-11 in
entirety. That said, there are bits and pieces of the manual pages for the object format
libraries that suggest there was some providence for PDP-11 in the development of COFF as
well.
Where this has landed though is a growing curiosity regarding:
- Whether SGS and COFF were tightly coupled to one another from the outset, with SGS being
supported by the general library routines being developed for the COFF format
- Whether COFF was envisioned as a one-size-fits-all object format from its inception or
started as an experiment in 3B-20 development that wound up being general enough for other
platforms
- If, prior to this format, there were any other efforts to produce a unifying binary
format and set of development tools, or if COFF was a happy accident from what were a
myriad of different architectural toolset streams
One of the curious things is how VAX for a brief moment did have its own set of tools and
a.out particulars before SGS/COFF. For instance, many of the VAX-targeted utilities in
3.0/System III bear little in common option/manual-wise with the general common SGS
utilities in System V. The "not on PDP-11" pages for various SGS components in
System V much more closely resemble the 3B-20 utilities in 4.1 than any of the non
PDP-11/VAX-only bits in System III.
Some examples:
- The VAX assembler in System III contains a -dN option indicating the number of bytes to
set aside for forward/external references for the linker to fill in.
- The VAX assembler in System V contains among others the -n and -m options from 4.1 which
indicate to disable address optimization and use m4 respectively
- The System V assembler goes on to also include -R (remove input file after completion)
-r (VAX only, add .data contents to .text instead) and options -b, -w, and -l to replace
the -d1, -d2, and -d4 options indicated in the previous VAX assembler
- System V further adds a -V to all the SGS software indicating the version of the
software. This is new circa 5.0, absent from the 4.1 manual like the R, r, b, w, and l
options
- The 4.1 manual's singular ar(1) entry still agrees with the System III version. No
arcv(1) is listed, implying the old ar format never made it to 3B-20
- The System V manual has both this ar(1) version as well as the new COFF-supporting
version. Not sure if this implies the VAX ar format was expanded to support the COFF stuff
for a little while until they decided on a new one or what.
- The System III ld (which is implied to support PDP and VAX) survives in System V, but is
cut down to supporting PDP-11 only
- The COFF-ish ld shows up in 4.1, is then extended to VAX presumably in the same breath
as the other COFF-supporting bits by Sys V, leading to two copies like many others,
PDP-11-specific stuff and then COFF-specific stuff
The picture that starts to form in the context of all of this is, for a little while in
the late 70s/early 80s, the software development environments for PDP-11, VAX-11, and
3B-20 were interplaying with each other in often times inconsistent ways. Taking a peek at
the 32V manuals, the VAX tools in System III appear to originate with that project, which
makes sense. If I'm understanding the timeline, COFF starts to emerge from the 3B-20
project and USG probably decides that's the way to go, a unified format, but with
PDP-11 pretty much out the door support wise already, there was little reason to apply
that to PDP-11 as well, so the PDP-11 tools get their swan song in System V, original
VAX-11 tools from 32V are likely killed off in 4.x, and the stuff that started with the
3B-20 group goes on to dominate the object file format and development software stuff
until ELF comes along some time later.
I guess other questions this raises are:
- Were the original VAX tools built with any attention to compatibility with the PDP-11
bits Ken and Dennis wrote many years prior (based on some option discrepancies, possibly
not?)
- Do the VAX utilities derive from the Interdata 8/32 work or if there was actually
another stream of tools as part of that project?
- Was there any interplay between the existing tool streams (original PDP-11, 32V's
VAX utilities, possibly Interdata 8/32) and the eventual COFF/SGS stuff, or was the latter
pretty well siloed in 3B-20 land until deployment with 4.1?
- Matt G.
23 Feb
23 Feb
4:20 a.m.
below are some thoughts/hopefully answers to your questions....
On Wed, Feb 22, 2023 at 3:16 PM segaloco via TUHS <tuhs(a)tuhs.org> wrote:
Good day all, figured I'd start a thread on this
matter as I'm starting to
piece enough together to articulate the questions arising in my research.
So based on my analysis of the 3B20S UNIX 4.1 manual I've been working
through, all evidence points to the formalized SGS package and COFF
originating tightly coupled to the 3B-20 line, then growing legs to support
VAX, but never quite absorbing PDP-11 in entirety. That said, there are
bits and pieces of the manual pages for the object format libraries that
suggest there was some providence for PDP-11 in the development of COFF as
well.
Where this has landed though is a growing curiosity regarding:
1. Whether SGS and COFF were tightly coupled to one another from the
outset, with SGS being supported by the general library routines being
developed for the COFF format
@scj - any enlightenment -- your team in USG must have been part of all
that.
1. Whether COFF was envisioned as a one-size-fits-all object format
from its inception or started as an experiment in 3B-20 development that
wound up being general enough for other platforms
That I can not say, but I can say that to the UNIX source licenses (i.e.
not the
Universities in the Research system or inside of the Bell Systems)
- it was used in the "consider it standard" campaign that AT&T marketing in
NC was starting to push. This was around the time that PCC2 was coming out
to replace the original PCC but I remember getting PCC2 was extra cost.
Most of the BSD based kernels (DEC, HP, etc..) were originally using a
modified a.out of their own flavor but I think almost all them switched to
COFF post the System III license. What I have forgotten, and it may have
been a requirement/mixed up in the license.
I do remember this was right around when gcc first starts coming out, and
they had a tool called robitussin to "cure coffs" as they were using a.out
wen they could.
1.
1. If, prior to this format, there were any other efforts to produce a
unifying binary format and set of development tools, or if COFF was a happy
accident from what were a myriad of different architectural toolset streams
MIT had a modified a.out format for the NU machine ports - that might
have been
called b.out.
CMU had macho which again was an extended a.out but even more flexible.
1. One of the curious things is how VAX for a brief moment did have
its own set of tools and a.out particulars before SGS/COFF.
Why is that curious - all original Vax development was just using the
original PCC
stream from V7 (and pre-Judge Green more in a minute).
What I don't remember is if PCC2 was COFF when introduced, or COFF can
first but I think they were separate things - again someone like scj would
be authoritative.
The three tools that have to care are the assembler (as), the linker (ld)
program loading code in the kernel itself.
For instance, many of the VAX-targeted utilities in
3.0/System III bear
little in common option/manual-wise with the general common SGS utilities
in System V. The "not on PDP-11" pages for various SGS components in System
V much more closely resemble the 3B-20 utilities in 4.1 than any of the non
PDP-11/VAX-only bits in System III.
Some examples:
- The VAX assembler in System III contains a -dN option indicating the
number of bytes to set aside for forward/external references for the linker
to fill in.
- The VAX assembler in System V contains among others the -n and -m
options from 4.1 which indicate to disable address optimization and use m4
respectively
- The System V assembler goes on to also include -R (remove input file
after completion) -r (VAX only, add .data contents to .text instead) and
options -b, -w, and -l to replace the -d1, -d2, and -d4 options indicated
in the previous VAX assembler
- System V further adds a -V to all the SGS software indicating the
version of the software. This is new circa 5.0, absent from the 4.1 manual
like the R, r, b, w, and l options
- The 4.1 manual's singular ar(1) entry still agrees with the System
III version. No arcv(1) is listed, implying the old ar format never made it
to 3B-20
Hmm this is confusing old v[456] ar format to new ar format was during
Research V6
to Research V7. By the time of any Vax development the old
format had pretty much been killed. I'd look at check what PWB 1.0 and 2.0
used. The new ar format was independent of what it was in it.
i.e. V7: man 5 ar
AR(5) <http://man.cat-v.org/unix_7th/5/AR>
AR(5) <http://man.cat-v.org/unix_7th/5/AR>
NAME
ar - archive (library) file format
SYNOPSIS
#include <ar.h>
DESCRIPTION
The archive command ar is used to combine several files into
one. Archives are used mainly as libraries to be searched
by the link-editor ld.
A file produced by ar has a magic number at the start, fol-
lowed by the constituent files, each preceded by a file
header. The magic number and header layout as described in
the include file are:
#define ARMAG 0177545
struct ar_hdr {
char ar_name[14];
long ar_date;
char ar_uid;
char ar_gid;
int ar_mode;
long ar_size;
};
- The System V manual has both this ar(1) version as well as the new
COFF-supporting version.
Why would ar(1) care?
- Not sure if this implies the VAX ar format was expanded to support
the COFF stuff for a little while until they decided on a new one or what.
- The System III ld (which is implied to support PDP and VAX) survives
in System V, but is cut down to supporting PDP-11 only
- The COFF-ish ld shows up in 4.1, is then extended to VAX presumably
in the same breath as the other COFF-supporting bits by Sys V, leading to
two copies like many others, PDP-11-specific stuff and then COFF-specific
stuff
The picture that starts to form in the context of all of this is, for a
little while in the late 70s/early 80s, the software development
environments for PDP-11, VAX-11, and 3B-20 were interplaying with each
other in often times inconsistent ways. Taking a peek at the 32V manuals,
the VAX tools in System III appear to originate with that project, which
makes sense. If I'm understanding the timeline, COFF starts to emerge from
the 3B-20 project and USG probably decides that's the way to go, a unified
format, but with PDP-11 pretty much out the door support wise already,
there was little reason to apply that to PDP-11 as well, so the PDP-11
tools get their swan song in System V, original VAX-11 tools from 32V are
likely killed off in 4.x, and the stuff that started with the 3B-20 group
goes on to dominate the object file format
That makes sense - but be careful - the 3B and WE32000 ISA may have been
the driver but I would expect that compiler folk in Summit were more in the
driver seat. The 3B20 kernel would use what they were getting from the
tools team and core kernel team in USG.
Remember the politic at the time is Judge Green has unleashed AT&T and they
are now allowed to be in the biz, and the sales/marketing folks AT&T was
pushing the 3B20 and the WE32000 - so there are big forces behind the
scenes that are not obvious/clear.
and development software stuff until ELF comes along
some time later.
Yep - never quite understood what the push for ELF was over COFF after all
the effort to drive COFF down people's throat. Note Microsoft "embraced
and extended" COFF as their format -- originally because of Xenix I believe.
Someone like Paul W may have some insights on this and that was before
the 3B20.
What was the format that the original Xenix used - when it was targeting
PDP-11, 68000, x86 and Z8000? Again I'm fuzzy on the details here. But I
do remember during the license discussions that would lead to System III,
that one of things the Microsoft team was worried about -- IIRC it was Bob
Greenberg pushing all that. I lost contact with Bob a few years ago, but
if we can find him, I would expect Bob to know what Xenix was doing. And
again that negotiation>>starts<< all pre-Judge Green, but finishes up soon
afterwards.
I guess other questions this raises are:
1. Were the original VAX tools built with any attention to
compatibility with the PDP-11 bits Ken and Dennis wrote many years prior
(based on some option discrepancies, possibly not?)
hrmph... folks started with the PDP-11 tools and changed them as needed.
I'm
not sure compatibility is the right term. They were retargeted nad
moved forward by people trying support a new machine they got and did not
want run DEC's OS.
1. Do the VAX utilities derive from the Interdata 8/32 work or if
there was actually another stream of tools as part of that project?
I guess I don't understand the question. The original V7 tools were
retargeted. When useful features were added, they might be
offered/returned to other folks, but remember, Research is not "supporting"
UNIX. USG is where things start to think in terms of multiple targets
>before Judge Green<< and then after Judge
Green, there was a push to stop
using non-AT&T based equipment or chips in the
Bell System and make what
Western Electric was selling be attractive [which sometimes was a little
bit of putting lipstick on porcine as it were]. For instance, Rob and
Barts's original JERQ is 68000 based, but by the time it becomes a product
as 5620 it has to be refactored as a WE32000.
1. Was there any interplay between the existing tool streams (original
PDP-11, 32V's VAX utilities, possibly Interdata 8/32) and the eventual
COFF/SGS stuff, or was the latter pretty well siloed in 3B-20 land until
deployment with 4.1?
I think you are putting too much on the 3B program itself. The 3B was the
task at
hand at the time and a solid opportunity to bring to bear business
choices being made. You need to look at the greater business to understand
a lot of the choices. A lot of things were happening in parallel in the
market that had other impacts on technology and how it was delivered -- the
3B program was the "technology train" leaving the station that some of them
got attached to/delivered using.
But, I as I said to you when we chatted, you really can not
underestimate what was happening (or not happening) as AT&T changed its
business focus - pre/post-Judge Green. It was a large company with lots of
different spheres of interest (read - different executives), each being
measured with different things that they might value.
6:17 a.m.
Thanks for the insights as always Clem! Mea culpa on not looking at the ar situation a
little more broadly, was pretty hyperfocused on 3B-20 stuff.
As for compatibility between development tools, I was mainly referring to available option
switches (the kind of thing that could potentially trip up someone's scripting). Once
SGS hits it seems like everything moving forward from that endeavored to keep comparable
command-line switches, but just to further compare:
- PDP as(1) uniquely has the '-' option to treat all undefined labels as globals
(as opposed to the much more common "this arg is stdin" that a lone
'-' would give elsewhere.) VAX as(1) drops this and adds -dN to define the size
reservation for undefined symbols. SGS as(1) keeps neither of these, opting to drop
'-' entirely and implement the -dN functionality as -b, -w, and -l instead.
- Things are a bit better for ld(1). SGS ld(1) drops the -X option of the earlier version
(pertains to cc behavior regarding internal labels, maybe irrelevant with pcc in the
picture). Also presumably the -n and -i options are dropped as their actions are already
default on VAX or otherwise only pertain to PDP. Old ld(1) had a -V option to store a
version string in the resulting object. This becomes -VS in SGS ld(1) to accommodate -V
being a standard "report my version" flag. SGS ld(1) then goes on to add -e to
explicitly denote the entry point, -f to provide a short int fill value for sections
needing it. We also pick up the now common -L for adding library paths. So all in all,
more commonality with pre-SGS ld(1) but still technically some breaking option changes.
- Looks like nm(1) may have some appreciable changes. The -g (only print globals), -p
(print in symbol table order), -r (print in reverse order), and -s (sort by size) option
values are removed. A few are replaced by different options: -n was originally sort
numerically instead of alphabetically (presumably by value rather than name), but in the
SGS version, this is reversed, -n being the print by name order option instead
(alphabetical is default in old nm(1)). The -o option morphs from meaning to include the
name of the source file in the output to print the symbol value as octal. For SGS nm(1),
we see the addition of -x (print in hex), -h (suppress headers), -v (sort by value,
presumably replaces the old -n meaning), -e (only print statics and externals), -f
("full" output), and -V (version). This presents breaking changes for all but
one of the switches to the earlier version of nm(1). For the record, V7, 32V, and System
III all appear to have a comparable version. This utility is particularly interesting
because a perusal of the current SUS
https://pubs.opengroup.org/onlinepubs/9699919799.2018edition/utilities/nm.h… shows a
mishmash of options, with -g and -u surviving all the way from V7, whereas -e, -f, -o, -v,
-x derive from the SGS behavior. Then there are the -A, -P, and -t options which are
explained in the rationale section of the standard; basically these are POSIX additions to
avoid using conflicting option types where possible.
- As for size(1) and strip(1), the SGS versions only add options.
- Finally, a few utilities are added. System III features a dump(1) command that is some
sort of tape dump utility, but this name repurposed at least as of 4.1 into an object file
section dumper, a role it retains. The list(1) utility is also added in 4.1.
So in detailed review, as(1), ld(1), and nm(1) had the most changes between the research
versions and what eventually landed in SGS, with nm(1) especially being wildly
incompatible from a command-line option standpoint. As for as(1), one PDP-11 option drops
and one VAX option changes switches (but not functionality as far as I can tell). Finally,
ld(1) seems to drop a few options that aren't needed in a non-PDP world and adjusts
the version-assignment option to allow -V to be a universal version request of the various
object utilities.
- Matt G.
------- Original Message -------
On Wednesday, February 22nd, 2023 at 2:20 PM, Clem Cole <clemc(a)ccc.com> wrote:
below are some thoughts/hopefully answers to your
questions....
On Wed, Feb 22, 2023 at 3:16 PM segaloco via TUHS <tuhs(a)tuhs.org> wrote:
Good day all, figured I'd start a thread on
this matter as I'm starting to piece enough together to articulate the questions
arising in my research.
So based on my analysis of the 3B20S UNIX 4.1 manual I've been working through, all
evidence points to the formalized SGS package and COFF originating tightly coupled to the
3B-20 line, then growing legs to support VAX, but never quite absorbing PDP-11 in
entirety. That said, there are bits and pieces of the manual pages for the object format
libraries that suggest there was some providence for PDP-11 in the development of COFF as
well.
Where this has landed though is a growing curiosity regarding:
- Whether SGS and COFF were tightly coupled to one another from the outset, with SGS
being supported by the general library routines being developed for the COFF format
@scj - any enlightenment -- your team in USG must have been part of all that.
- Whether COFF was envisioned as a
one-size-fits-all object format from its inception or started as an experiment in 3B-20
development that wound up being general enough for other platforms
That I can not say, but I can say that to the UNIX source licenses (i.e. not the
Universities in the Research system or inside of the Bell Systems) - it was used in the
"consider it standard" campaign that AT&T marketing in NC was starting to
push. This was around the time that PCC2 was coming out to replace the original PCC but I
remember getting PCC2 was extra cost.
Most of the BSD based kernels (DEC, HP, etc..) were originally using a modified a.out of
their own flavor but I think almost all them switched to COFF post the System III license.
What I have forgotten, and it may have been a requirement/mixed up in the license.
I do remember this was right around when gcc first starts coming out, and they had a tool
called robitussin to "cure coffs" as they were using a.out wen they could.
-
- If, prior to this format, there were any other efforts to produce a unifying binary
format and set of development tools, or if COFF was a happy accident from what were a
myriad of different architectural toolset streams
MIT had a modified a.out format for the NU machine ports - that might have been called
b.out.
CMU had macho which again was an extended a.out but even more flexible.
- One of the curious things is how VAX for a
brief moment did have its own set of tools and a.out particulars before SGS/COFF.
Why is that curious - all original Vax development was just using the original PCC stream
from V7 (and pre-Judge Green more in a minute).
What I don't remember is if PCC2 was COFF when introduced, or COFF can first but I
think they were separate things - again someone like scj would be authoritative.
The three tools that have to care are the assembler (as), the linker (ld) program loading
code in the kernel itself.
For instance, many of the VAX-targeted utilities
in 3.0/System III bear little in common option/manual-wise with the general common SGS
utilities in System V. The "not on PDP-11" pages for various SGS components in
System V much more closely resemble the 3B-20 utilities in 4.1 than any of the non
PDP-11/VAX-only bits in System III.
Some examples:
- The VAX assembler in System III contains a -dN option indicating the number of bytes to
set aside for forward/external references for the linker to fill in.
- The VAX assembler in System V contains among others the -n and -m options from 4.1
which indicate to disable address optimization and use m4 respectively
- The System V assembler goes on to also include -R (remove input file after completion)
-r (VAX only, add .data contents to .text instead) and options -b, -w, and -l to replace
the -d1, -d2, and -d4 options indicated in the previous VAX assembler
- System V further adds a -V to all the SGS software indicating the version of the
software. This is new circa 5.0, absent from the 4.1 manual like the R, r, b, w, and l
options
- The 4.1 manual's singular ar(1) entry still agrees with the System III version. No
arcv(1) is listed, implying the old ar format never made it to 3B-20
Hmm this is confusing old v[456] ar format to new ar format was during Research V6 to
Research V7. By the time of any Vax development the old format had pretty much been
killed. I'd look at check what PWB 1.0 and 2.0 used. The new ar format was
independent of what it was in it.
i.e. V7: man 5 ar
[AR(5)](http://man.cat-v.org/unix_7th/5/AR)
[AR(5)](http://man.cat-v.org/unix_7th/5/AR)
NAME
ar - archive (library) file format
SYNOPSIS
#include <ar.h>
DESCRIPTION
The archive command ar is used to combine several files into
one. Archives are used mainly as libraries to be searched
by the link-editor ld.
A file produced by ar has a magic number at the start, fol-
lowed by the constituent files, each preceded by a file
header. The magic number and header layout as described in
the include file are:
#define
ARMAG 0177545
> struct ar_hdr { > char
ar_name[14];
> long ar_date;
> char ar_uid;
> char ar_gid;
> int ar_mode;
>> long ar_size; > };
- The System V manual has both this ar(1) version
as well as the new COFF-supporting version.
Why would ar(1) care?
- Not sure if this implies the VAX ar format was
expanded to support the COFF stuff for a little while until they decided on a new one or
what.
- The System III ld (which is implied to support PDP and VAX) survives in System V, but
is cut down to supporting PDP-11 only
- The COFF-ish ld shows up in 4.1, is then extended to VAX presumably in the same breath
as the other COFF-supporting bits by Sys V, leading to two copies like many others,
PDP-11-specific stuff and then COFF-specific stuff
The picture that starts to form in the context of all of this is, for a little while in
the late 70s/early 80s, the software development environments for PDP-11, VAX-11, and
3B-20 were interplaying with each other in often times inconsistent ways. Taking a peek at
the 32V manuals, the VAX tools in System III appear to originate with that project, which
makes sense. If I'm understanding the timeline, COFF starts to emerge from the 3B-20
project and USG probably decides that's the way to go, a unified format, but with
PDP-11 pretty much out the door support wise already, there was little reason to apply
that to PDP-11 as well, so the PDP-11 tools get their swan song in System V, original
VAX-11 tools from 32V are likely killed off in 4.x, and the stuff that started with the
3B-20 group goes on to dominate the object file format
That makes sense - but be careful - the 3B and WE32000 ISA may have been the driver but I
would expect that compiler folk in Summit were more in the driver seat. The 3B20 kernel
would use what they were getting from the tools team and core kernel team in USG.
Remember the politic at the time is Judge Green has unleashed AT&T and they are now
allowed to be in the biz, and the sales/marketing folks AT&T was pushing the 3B20 and
the WE32000 - so there are big forces behind the scenes that are not obvious/clear.
and development software stuff until ELF comes
along some time later.
Yep - never quite understood what the push for ELF was over COFF after all the effort to
drive COFF down people's throat. Note Microsoft "embraced and extended"
COFF as their format -- originally because of Xenix I believe. Someone like Paul W may
have some insights on this and that was before the 3B20.
What was the format that the original Xenix used - when it was targeting PDP-11, 68000,
x86 and Z8000? Again I'm fuzzy on the details here. But I do remember during the
license discussions that would lead to System III, that one of things the Microsoft team
was worried about -- IIRC it was Bob Greenberg pushing all that. I lost contact with Bob
a few years ago, but if we can find him, I would expect Bob to know what Xenix was doing.
And again that negotiation>>starts<< all pre-Judge Green, but finishes up soon
afterwards.
I guess other questions this raises are:
- Were the original VAX tools built with any attention to compatibility with the PDP-11
bits Ken and Dennis wrote many years prior (based on some option discrepancies, possibly
not?)
hrmph... folks started with the PDP-11 tools and changed them as needed. I'm not
sure compatibility is the right term. They were retargeted nad moved forward by people
trying support a new machine they got and did not want run DEC's OS.
- Do the VAX utilities derive from the Interdata
8/32 work or if there was actually another stream of tools as part of that project?
I guess I don't understand the question. The original V7 tools were retargeted. When
useful features were added, they might be offered/returned to other folks, but remember,
Research is not "supporting" UNIX. USG is where things start to think in terms
of multiple targets >>before Judge Green<< and then after Judge Green, there
was a push to stop using non-AT&T based equipment or chips in the Bell System and make
what Western Electric was selling be attractive [which sometimes was a little bit of
putting lipstick on porcine as it were]. For instance, Rob and Barts's original JERQ
is 68000 based, but by the time it becomes a product as 5620 it has to be refactored as a
WE32000.
- Was there any interplay between the existing
tool streams (original PDP-11, 32V's VAX utilities, possibly Interdata 8/32) and the
eventual COFF/SGS stuff, or was the latter pretty well siloed in 3B-20 land until
deployment with 4.1?
I think you are putting too much on the 3B program itself. The 3B was the task at hand at
the time and a solid opportunity to bring to bear business choices being made. You need to
look at the greater business to understand a lot of the choices. A lot of things were
happening in parallel in the market that had other impacts on technology and how it was
delivered -- the 3B program was the "technology train" leaving the station that
some of them got attached to/delivered using.
But, I as I said to you when we chatted, you really can not underestimate what was
happening (or not happening) as AT&T changed its business focus - pre/post-Judge
Green. It was a large company with lots of different spheres of interest (read - different
executives), each being measured with different things that they might value. 
12:30 p.m.
Clem Cole wrote:
MIT had a modified a.out format for the NU machine
ports - that might
have been called b.out. CMU had macho which again was an extended
a.out but even more flexible.
Digital Research's GEMDOS also used a modified a.out format (at least as
found on the Atari ST), pretty much the original PDP-11 format with
32-bit addresses.
[Shouldn't this go to the COFF mailing list... oh sorry.]
8:25 p.m.
Point of interest. In the mid 80's Northern Telecom had a product called
the DV-1. It was a
motorola based system on the 68010. At the end of its life it was able to
run GEM with a floppy.
The DV-1 terminal was 68000 based. See some snaps at:
https://www.flickr.com/photos/9479603@N02/1814557731/in/album-7215760282421…
https://www.flickr.com/photos/9479603@N02/1814548075/in/album-7215760282421…
https://www.flickr.com/photos/9479603@N02/1814557771/in/album-7215760282421…
https://www.flickr.com/photos/9479603@N02/1814557653/in/album-7215760282421…
https://www.flickr.com/photos/9479603@N02/albums/72157602824219250/page2
My complete photo/movie history is at:
https://www.flickr.com/photos/9479603@N02/albums
If you have any to add please pass them along.
Ken
On Thu, Feb 23, 2023 at 1:30 AM Lars Brinkhoff <lars(a)nocrew.org> wrote:
Clem Cole wrote:
--
End of line
JOB TERMINATED
MIT had a modified a.out format for the NU
machine ports - that might
have been called b.out. CMU had macho which again was an extended
a.out but even more flexible.
Digital Research's GEMDOS also used a modified a.out format (at least as
found on the Atari ST), pretty much the original PDP-11 format with
32-bit addresses.
[Shouldn't this go to the COFF mailing list... oh sorry.]
24 Feb
24 Feb
1:37 a.m.
On Wed, Feb 22, 2023 at 11:30 PM Lars Brinkhoff <lars(a)nocrew.org> wrote:
Clem Cole wrote:
MIT's PCC ports to various micros, especailly x86, also used b.out that was
quite similar to a.out.
Warner
MIT had a modified a.out format for the NU
machine ports - that might
have been called b.out. CMU had macho which again was an extended
a.out but even more flexible.
Digital Research's GEMDOS also used a modified a.out format (at least as
found on the Atari ST), pretty much the original PDP-11 format with
32-bit addresses.
11:01 p.m.
MIT's PCC ports to various micros, especailly
x86, also used b.out that
was quite similar to a.out.
That was a cross-compiler, right? If so, I remember going over to MIT-LCS
and picking up a copy of the sources for the compiler and the MIT PC/IP
implementation. I had to show a copy of my ATT source license to the
secretary, who made a copy of a 9track tape for me.She was a hoot: dressed
like a punk, and we takled about not working for defense companies. We were
then able to build and install various TCP/IP utilities on our PC-AT
machines running early Windows releases. This was probably early 1980's. I
think the PC/IP code was based on David Clark's upcalls idea[1]. It later
became the basis for FTP software, which then died when Microsoft added TCP
to windows.
[1]
https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=e7c6…
23 Feb
23 Feb
10:49 p.m.
On 2/22/23, Clem Cole <clemc(a)ccc.com> wrote:
a.out was, as object file formats go, a throwback to the stone age
from the get-go. Even the most primitive of IBM's link editors for
System/360 supported arbitrary naming of object file sections and the
ability for the programmer to arrange them in whatever order they
wished. a.out's restriction to three sections (.text, .data, .bss)
did manage to get the job done, and even (with ZMAGIC) could support
demand-paged virtual memory, but only just.
It became pretty clear in the 1980s that an object file format more
powerful and flexible than a.out was needed. CMU developed their own
object file format (MACH-O) for their MACH microkernel-based OS. It
had up to 8 object file sections, and the section properties (e.g.,
read vs. read/wrkte; executable vs. data) were not tied to the section
name as in a.out. A big step forward, although still primitive
compared to the object formats of VAX/VMS and the IBM S/370 OSes.
Apple MacOS X still uses MACH-O for object files and executables.
Whatever its origins, what we now know as COFF (Common Object File
Format) is, as its name implies, intended to be OS- and
machine-independent. It still has a relatively small number of
sections, albeit more than MACH-O. When Microsoft developed Windows
NT, they needed to replace their own MZ executable format with
something that could support shareable images and they decided to go
with COFF for both object files and for executables. In typical
Microsoft embrace-and-extend fashion, their Portable Executable and
Common Object File Format (PECOFF) is a heavily modified version of
COFF with lots of MS-specific extensions. When DEC's GEM back end was
chosen as the optimizer and code generator for Microsoft C/C++ on
Windows NT for the DEC Alpha chip, I had to add PECOFF support to
GEM's existing COFF support (which was used by DEC's commercially sold
compilers for Ultrix). My original idea was to put the PECOFF support
under conditional compilation (#ifdef PECOFF), but the two formats
were sufficiently different that I abandoned that Idea, cloned the
existing COFF module, and then modified that to create a separate
PECOFF module.
ELF is far more flexible than either COFF, PECOFF, or MACH-O. Those
three make a distinction between sections (the bits that eventually
end up in memory) and the metadata pieces of an object file or
executable (program headers, symbol table, debug information, etc.).
In ELF, everything is a section, including the symbol table and the
tables that direct the program loader in mapping shareable images into
a process's memory. ELF was originally limited to 64K sections
(section numbers were unsigned 16-bit), but there is now a scheme for
32-bit section numbers. The essentially unlimited number of sections
is a big boon to languages such as C++, where grouped sections with a
name-decoration convention provide a convenient way to support sharing
of class definitions without requiring language-specific tweaks to the
software development toolset. Contrast this with the Ada
implementations I'm aware of, which have their own software
development library systems layered on top of the conventional
compiler/linker/archiver to insure that program modules are compiled
and linked in the correct order.
I don't know what the timeline for the invention of COFF was. It was
already called COFF and in widespread use by the time I encountered it
when we added Ultrix support to GEM. I think MACH-O predated COFF;
it's certainly more primitive than COFF. MACH-O was probably early to
mid-1980s. OS kernel bloat was a recognized problem at the time and
microkernel-based OSes were all the rage. At DEC, Dave Cutler wrote a
microkernel-based OS called VAXeln to replace VAX/VMS for real-time
applications. A lot of concepts from VAXeln found their way into
Windows NT when Cutler left DEC for Microsoft.
-Paul W.
>
> - The System V manual has both this ar(1) version as well as the new
> COFF-supporting version.
>
> Why would ar(1) care?
>
> - Not sure if this implies the VAX ar format was expanded to support
> the COFF stuff for a little while until they decided on a new one or
> what.
I can't think of any reason why ar(1) would care about the file format
or internal contents of any of the modules it archives. ar(1) is a
general archiving tool and can archive anything. It happens that the
designers of ld(1) decided to use ar(1) to provide searchable object
file libraries.
ranlib(1) is a different matter. In order to index global symbols it
has to understand the object file format(s) of the modules it is
indexing. ranlib(1) most certainly would have to be taught to
understand COFF. But not ar(1).
and
development software stuff until ELF comes along some time later.
Yep - never quite understood what the push for ELF was over COFF after all
the effort to drive COFF down people's throat. Note Microsoft "embraced
and extended" COFF as their format -- originally because of Xenix I
believe.
Someone like Paul W may have some insights on this and that was before
the 3B20.
24 Feb
24 Feb
12:38 a.m.
For the sake of timelines:
June 1980 - Publication date on the front page of the 3.0 manual in which the utilities
are still very much research for PDP-11 and 32V-ish for VAX where distinctions matter.
June 1981 - Publication date on the front page of the 4.1 manual in which the man-pages
very much refer to all of this as the "3B-20 object format"
June 1982 - Publication date on the front page of the 5.0 manual by which point these same
pages had been edited and extended to describe the "common object file format"
Additions at the 1981 release include dump(1), list(1), and the ld-prefixed library
routines for managing these object files. These likewise persist in 5.0, SysV, and beyond
as COFF-related tools.
So this puts the backstop of what would become COFF at at least '81.
- Matt G.
------- Original Message -------
On Thursday, February 23rd, 2023 at 8:49 AM, Paul Winalski <paul.winalski(a)gmail.com>
wrote:
On 2/22/23, Clem Cole clemc(a)ccc.com wrote:
a.out was, as object file formats go, a throwback to the stone age
from the get-go. Even the most primitive of IBM's link editors for
System/360 supported arbitrary naming of object file sections and the
ability for the programmer to arrange them in whatever order they
wished. a.out's restriction to three sections (.text, .data, .bss)
did manage to get the job done, and even (with ZMAGIC) could support
demand-paged virtual memory, but only just.
It became pretty clear in the 1980s that an object file format more
powerful and flexible than a.out was needed. CMU developed their own
object file format (MACH-O) for their MACH microkernel-based OS. It
had up to 8 object file sections, and the section properties (e.g.,
read vs. read/wrkte; executable vs. data) were not tied to the section
name as in a.out. A big step forward, although still primitive
compared to the object formats of VAX/VMS and the IBM S/370 OSes.
Apple MacOS X still uses MACH-O for object files and executables.
Whatever its origins, what we now know as COFF (Common Object File
Format) is, as its name implies, intended to be OS- and
machine-independent. It still has a relatively small number of
sections, albeit more than MACH-O. When Microsoft developed Windows
NT, they needed to replace their own MZ executable format with
something that could support shareable images and they decided to go
with COFF for both object files and for executables. In typical
Microsoft embrace-and-extend fashion, their Portable Executable and
Common Object File Format (PECOFF) is a heavily modified version of
COFF with lots of MS-specific extensions. When DEC's GEM back end was
chosen as the optimizer and code generator for Microsoft C/C++ on
Windows NT for the DEC Alpha chip, I had to add PECOFF support to
GEM's existing COFF support (which was used by DEC's commercially sold
compilers for Ultrix). My original idea was to put the PECOFF support
under conditional compilation (#ifdef PECOFF), but the two formats
were sufficiently different that I abandoned that Idea, cloned the
existing COFF module, and then modified that to create a separate
PECOFF module.
ELF is far more flexible than either COFF, PECOFF, or MACH-O. Those
three make a distinction between sections (the bits that eventually
end up in memory) and the metadata pieces of an object file or
executable (program headers, symbol table, debug information, etc.).
In ELF, everything is a section, including the symbol table and the
tables that direct the program loader in mapping shareable images into
a process's memory. ELF was originally limited to 64K sections
(section numbers were unsigned 16-bit), but there is now a scheme for
32-bit section numbers. The essentially unlimited number of sections
is a big boon to languages such as C++, where grouped sections with a
name-decoration convention provide a convenient way to support sharing
of class definitions without requiring language-specific tweaks to the
software development toolset. Contrast this with the Ada
implementations I'm aware of, which have their own software
development library systems layered on top of the conventional
compiler/linker/archiver to insure that program modules are compiled
and linked in the correct order.
I don't know what the timeline for the invention of COFF was. It was
already called COFF and in widespread use by the time I encountered it
when we added Ultrix support to GEM. I think MACH-O predated COFF;
it's certainly more primitive than COFF. MACH-O was probably early to
mid-1980s. OS kernel bloat was a recognized problem at the time and
microkernel-based OSes were all the rage. At DEC, Dave Cutler wrote a
microkernel-based OS called VAXeln to replace VAX/VMS for real-time
applications. A lot of concepts from VAXeln found their way into
Windows NT when Cutler left DEC for Microsoft.
-Paul W.
> - The System V manual has both this ar(1)
version as well as the new
> COFF-supporting version.
>
> Why would ar(1) care?
>
> - Not sure if this implies the VAX ar format was expanded to support
> the COFF stuff for a little while until they decided on a new one or
> what.
I can't think of any reason why ar(1) would care about the file format
or internal contents of any of the modules it archives. ar(1) is a
general archiving tool and can archive anything. It happens that the
designers of ld(1) decided to use ar(1) to provide searchable object
file libraries.
ranlib(1) is a different matter. In order to index global symbols it
has to understand the object file format(s) of the modules it is
indexing. ranlib(1) most certainly would have to be taught to
understand COFF. But not ar(1).
and
development software stuff until ELF comes along some time later.
Yep - never quite understood what the push for ELF was over COFF after all
the effort to drive COFF down people's throat. Note Microsoft "embraced
and extended" COFF as their format -- originally because of Xenix I
believe.
Someone like Paul W may have some insights on this and that was before
the 3B20.
2:40 a.m.
One property of a.out is that the format of an executable image is
identical to that of an object file. One could write an assembly
program that was self-contained and did not need to be linked to any
other modules, run that through as(1). The resulting a.out file would
be executable. No need to involve ld(1).
MACH-O retained this concept of supporting self-contained object files
as executables. COFF departed from that tradition. A COFF (or
PECOFF) executable requires a data structure in the file called the
optional header that contains important instructions to the program
loader. This second header is "optional" only in the sense that
non-executable object files do not have one. The optional header is
created by ld(1). I'm almost certain that versions of as(1) that
generate COFF output do not generate an optional header.
In ELF, executable images have several sections such as the program
header table section that contain the information present in COFF's
optional header.
Traditionally, operating systems such as IBM's OS for System/360/370/Z
and DEC's VMS have very different formats for object files and
executable programs. OpenVMS for Itanium and x86-64 use ELF for both
objects and executables--a departure from the practice used on VAX and
Alpha.
-Paul W.
6:45 p.m.
Paul Winalski <paul.winalski(a)gmail.com> wrote:
I can't think of any reason why ar(1) would care
about the file format
or internal contents of any of the modules it archives. ar(1) is a
general archiving tool and can archive anything. It happens that the
designers of ld(1) decided to use ar(1) to provide searchable object
file libraries.
ranlib(1) is a different matter. In order to index global symbols it
has to understand the object file format(s) of the modules it is
indexing. ranlib(1) most certainly would have to be taught to
understand COFF. But not ar(1).
You are correct that ar(1) was originally just an archiver. However,
the System V people built ranlib into it; the .a file for a library
has a sort of hidden extra member that is the list of symbols in
the archive.
With tar and cpio, ar apparently fell out of use as a general
archiver, and today it's only used for libraries of relocatable
object files.
Arnold
7:13 p.m.
On Fri, Feb 24, 2023 at 05:45:16AM -0700, arnold(a)skeeve.com wrote:
With tar and cpio, ar apparently fell out of use as a
general
archiver, and today it's only used for libraries of relocatable
object files.
'ar' is alive and well as the archive format for .deb files though, so it
could be argued that it's actually hugely in use as an archiver all across
the world :)
Bye, Arno.
26 Feb
26 Feb
1:28 a.m.
Arno Griffioen via TUHS <tuhs(a)tuhs.org> wrote:
On Fri, Feb 24, 2023 at 05:45:16AM -0700,
arnold(a)skeeve.com wrote:
I wasn't aware of that. Thanks.
With tar and cpio, ar apparently fell out of use
as a general
archiver, and today it's only used for libraries of relocatable
object files.
'ar' is alive and well as the archive format for .deb files though, so it
could be argued that it's actually hugely in use as an archiver all across
the world :)
Bye, Arno. 
1:34 a.m.
arnold(a)skeeve.com wrote in
<202302251928.31PJSXc1004140(a)freefriends.org>:
|Arno Griffioen via TUHS <tuhs(a)tuhs.org> wrote:
|> On Fri, Feb 24, 2023 at 05:45:16AM -0700, arnold(a)skeeve.com wrote:
|>> With tar and cpio, ar apparently fell out of use as a general
|>> archiver, and today it's only used for libraries of relocatable
|>> object files.
|>
|> 'ar' is alive and well as the archive format for .deb files though, \
|> so it
|> could be argued that it's actually hugely in use as an archiver all \
|> across
|> the world :)
|>
|> Bye, Arno.
|
|I wasn't aware of that. Thanks.
It is also used during the build phase of some programs (groff
last i looked). Ie they build their support objects, and instead
of having make rules with $(OBJ) they simply link against the
library. (I have not looked whether they build with -fPIC
/ -fPIE, to outrule your "relocatable", that is.)
--End of <202302251928.31PJSXc1004140(a)freefriends.org>
--steffen
|
|Der Kragenbaer, The moon bear,
|der holt sich munter he cheerfully and one by one
|einen nach dem anderen runter wa.ks himself off
|(By Robert Gernhardt)
25 Feb
25 Feb
8:07 a.m.
On Thu, 23 Feb 2023, Paul Winalski wrote:
a.out was, as object file formats go, a throwback to
the stone age from
the get-go. Even the most primitive of IBM's link editors for
System/360 supported arbitrary naming of object file sections and the
ability for the programmer to arrange them in whatever order they
wished. a.out's restriction to three sections (.text, .data, .bss) did
manage to get the job done, and even (with ZMAGIC) could support
demand-paged virtual memory, but only just.
That may be so, but those guys didn't exactly have the resources of
IBM behind them...
And I wonder how many people here know the significance of the "407" magic
number?
-- Dave
9:30 p.m.
On Fri, Feb 24, 2023 at 9:07 PM Dave Horsfall <dave(a)horsfall.org> wrote:
On Thu, 23 Feb 2023, Paul Winalski wrote:
A reasonable point, but it was more Ken, Dennis, and the team did what
a.out was, as object file formats go, a throwback
to the stone age from
the get-go. Even the most primitive of IBM's link editors for
System/360 supported arbitrary naming of object file sections and the
ability for the programmer to arrange them in whatever order they
wished. a.out's restriction to three sections (.text, .data, .bss) did
manage to get the job done, and even (with ZMAGIC) could support
demand-paged virtual memory, but only just.
That may be so, but those guys didn't exactly have the resources of
IBM behind them...
>they needed<< and it was good enough for a
long time. The IBM link
editors needed all that back in the day. As more and more
"modern"
languages came into being, it was not until about 6th editions that
difficulties of not having an expandable object format and better linker
began to show, and as Paul says, until the support for demand paging that
a.out was really stressed.
Please correct me if I'm misinformed, but Paul, of course, had to support
the DEC language tools on Unix, which had come from systems that had a more
flexible format (the solution for Ultrix IICR was to move a flavor of the
VMS linker to UNIX for object file and just a.out for execution). So he
lived the difficulties/shortcomings. A valid argument is Tanndenbaum's
compiler toolkit survived with a.out, and he supported many of the same
language targets that DEC did. Andy and crew did their own assemblers,
does anyone remember if they supplied a new linker and object format? That
would make Paul's point more powerful -> the languages people wanted
something more.
My point, the UNIX developers built what they needed and built that well.
Their format worked with their development primary language/tool (a.k.a. C)
and they even made it work with an f77 implementation. So it is a little
hard to knock them too much -- it was not part of the original design spec.
But as Matt has discussed in his digging through things, it does look like
as the AT&T languages team started to run into some of the same barriers,
they started to move to a new format.
And I wonder how many people here know the significance of the "407" magic
number?
Today, few and fewer I fear. For those do not, please see page 4-33 of the
1975/76 DEC PDP-11 Processor Handbook and think about boot blocks. 🍺
ᐧ
11:29 p.m.
On 2/25/23, Clem Cole <clemc(a)ccc.com> wrote:
The IBM link
editors needed all that back in the day.
One of the complications of executable images was the address space
layout for OS/360. There was no virtual memory. The low part of the
address space was where the operating system kernel (supervisor, in
IBM-speak) lived. Each of what we now call processes was assigned a
contiguous portion of the remaining address space (this was called a
partition). As executable program retained the relocations that had
been applied by the linker so that the program loader could adjust the
addresses in the executable depending on which partition it was being
loaded into. This made things more complicated for the link editor.
The OS/360 link editor also doubled as a patch tool. The link editor
was capable of taking an executable and a set of new versions for some
object modules, un-linking those modules, and replacing them with the
new versions, adjusting all relocations accordingly.
Please correct me if I'm misinformed, but Paul,
of course, had to support
the DEC language tools on Unix, which had come from systems that had a more
flexible format (the solution for Ultrix IICR was to move a flavor of the
VMS linker to UNIX for object file and just a.out for execution).
VAX Fortran for Ultrix was a port of the VAX/VMS Fortran compiler and
runtime to Ultrix. There were two big problems, object-file-wise.
First, VAX Fortran used many of the advanced features of the VMS
object language. To generate a.out directly, those chores would have
to be done in the compiler's code generator. The runtime library had
an even worse problem. One innovative feature of VMS was that it had
a very robust ABI that could support every programming language then
known, and the compiler development teams were required to adhere to
this ABI and to provide language extensions were necessary to support
all of the ABI's features. The result was that calling subroutines
written in a different language was dead easy. It didn't matter which
programming language you used. The developers of the Fortran runtime
took full advantage of that, and it contained routines written in
several languages. This meant that we would either need to add a.out
support to the code generators of all of those compilers (there was no
common back end in those days) or write a VMS .obj-to-a.out
translator. In the end we decided that the easiest solution to both
problems was to add a.out support to the VMS linker and to port it to
Ultrix. The result was lk(1), a linker that accepted either VMS .obj
or a.out as input and generated an a.out executable.
My point, the UNIX developers built what they needed
and built that well.
Amen to that. That observation applies to a lot of the design of Unix.
Their format worked with their development primary
language/tool (a.k.a. C)
and they even made it work with an f77 implementation. So it is a little
hard to knock them too much -- it was not part of the original design spec.
They did what was right given their circumstances and resources. I
know of two major operating system development efforts at DEC, OFIS
and MICA, that illustrate what happens if you try to take the opposite
tack. Their thinking was, "Gee, we have a lot to do and a short
schedule. I know--let's invent a new programming language and write a
compiler for it." Both ended up being disastrous examples of Second
System Syndrome.
And I wonder how many people here know the
significance of the "407" magic
Cute.
-Paul W.
number?
Today, few and fewer I fear. For those do not, please see page 4-33 of the
1975/76 DEC PDP-11 Processor Handbook and think about boot blocks. 🍺
ᐧ
666
days inactive
669
days old
17 comments
12 participants
participants (12)
-
Arno Griffioen -
arnold＠skeeve.com -
Clem Cole -
Dave Horsfall -
Harald Arnesen -
KenUnix -
Lars Brinkhoff -
Paul Winalski -
Rich Salz -
segaloco -
Steffen Nurpmeso -
Warner Losh