Retrocomputing - Forth for PDP-7

[MitchBradley]

Inspired by a friend who is working on a PDP-7 (the computer on which Unix was first developed) emulator, I asked Claude to help me develop a PDP-7 Forth with Turtle Graphics . It fits in 2K of the machine's 8K 18-bit words, including stacks and buffers, with an additional 1K words wired down for the graphics buffer. It is fairly complete as old-school Forths go, but I didn't bother to implement CREATE .. DOES> . The thread format is native machine instructions - calls to primitives are JMP

instructions, colon defs are called by CAL , constants and literals are LAC , and variables are LAW . To save space, it uses the ancient 3-characters+count header format, with each header taking exactly 2 words.

FIND is 22 instructions. NEXT is 1 instruction in the usual case, and 3 for constants and variables. Pushing onto the stack is cheap, but popping takes 6 instructions due to the way that auto-incrementing (but not auto-decrementing) index registers work. BRANCH and ?BRANCH are pretty conventional for a threaded code Forth - the branch offset follows the primitive in the body.

The turtle graphics package is 113 lines of Forth code, compiling to 700 words.

You can load files from emulated paper tape. You can run conditionals and loops directly from interpret state using the temporary compilation technique that I developed for Open Firmware. Here is a Turtle Graphics drawing and its code:

: SQ 4 0 DO 200 FD 90 RT LOOP ;  OK
8 0 DO SQ 45 RT LOOP  OK

[SVFIG-Kevin]

That sounds like a great presentation for the monthly SVFIG Zoom meeting.   If you don't want to do a full-up presentation, perhaps a 10 or 15 minute overview on Forth Day. Anyone who wishes to present to SVFIG is welcome to contact me at ***@***.***

…

On 9/26/2026 7:01 PM, Mitch Bradley wrote: Inspired by a friend who is working on a PDP-7 (the computer on which Unix was first developed) emulator, I asked Claude to help me develop a PDP-7 Forth with Turtle Graphics <https://MitchBradley/pdp7forth> . It fits in 2K of the machine's 8K 18-bit words, including stacks and buffers, with an additional 1K words wired down for the graphics buffer. It is fairly complete as old-school Forths go, but I didn't bother to implement CREATE .. DOES> . The thread format is native machine instructions - calls to primitives are JMP instructions, colon defs are called by CAL , constants and literals are LAC , and variables are LAW . To save space, it uses the ancient 3-characters+count header format, with each header taking exactly 2 words. FIND is 22 instructions. NEXT is 1 instruction in the usual case, and 3 for constants and variables. Pushing onto the stack is cheap, but popping takes 6 instructions due to the way that auto-incrementing (but not auto-decrementing) index registers work. BRANCH and ?BRANCH are pretty conventional for a threaded code Forth - the branch offset follows the primitive in the body. The turtle graphics package is 113 lines of Forth code, compiling to 700 words. You can load files from emulated paper tape. You can run conditionals and loops directly from interpret state using the temporary compilation technique that I developed for Open Firmware. Here is a Turtle Graphics drawing and its code: |: SQ 4 0 DO 200 FD 90 RT LOOP ; OK 8 0 DO SQ 45 RT LOOP OK | Image: image (view on web) <https://github.com/user-attachments/assets/55b55bce-5294-4b59-a55b-c591a5e0e3a3> — Reply to this email directly, view it on GitHub <#206?email_source=notifications&email_token=ADUZY7BLZOUHSQMWJJVAUDL5RBYJFA5CNFSNUABBM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63RPGEYDQOJWG4YDHJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KYZTPN52GK4S7MNWGSY3L>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADUZY7FJYFZVUDFZ5RL2G5T5RBYJFAVCNFSNUABHKJSXA33TNF2G64TZHMZTCNBRGQ2DCNR3IRUXGY3VONZWS33OHMYTAOBZGY3TAM5BOYBA>. Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS <https://github.com/notifications/mobile/ios/ADUZY7GREQKHEMAV27UHUMD5RBYJFA5CNFSNUABBM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63RPGEYDQOJWG4YDHJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KUZTPN52GK4S7NFXXG> and Android <https://github.com/notifications/mobile/android/ADUZY7DT44KJ3JCB4LEGU7L5RBYJFA5CNFSNUABBM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63RPGEYDQOJWG4YDHJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2K4ZTPN52GK4S7MFXGI4TPNFSA>. Download it today! You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[MitchBradley]

When is forth day this year? On Sep 27, 2026, at 11:48 AM, Kevin Appert ***@***.***> wrote: That sounds like a great presentation for the monthly SVFIG Zoom meeting.   If you don't want to do a full-up presentation, perhaps a 10 or 15 minute overview on Forth Day. Anyone who wishes to present to SVFIG is welcome to contact me at ***@***.***

[SVFIG-Kevin]

By long tradition, Forth Day is the third Saturday of November.  This year, that falls on 21 November 2026. December is also the third Saturday.  All the rest of the year the monthly meetings are on the fourth Saturday. forther at comcast dot net

…

On 9/27/2026 7:02 PM, Mitch Bradley wrote: When is forth day this yearOn Sep 27, 2026, at 11:48 AM, Kevin Appert ***@***.***> wrote: That sounds like a great presentation for the monthly SVFIG Zoom meeting.   If you don't want to do a full-up presentation, perhaps a 10 or 15 minute overview on Forth Day. Anyone who wishes to present to SVFIG is welcome to contact me at ***@***.*** On 9/26/2026 7:01 PM, Mitch Bradley wrote: > > Inspired by a friend who is working on a PDP-7 (the computer on which > Unix was first developed) emulator, I asked Claude to help me develop > a PDP-7 Forth with Turtle Graphics <https://MitchBradley/pdp7forth> . > It fits in 2K of the machine's 8K 18-bit words, including stacks and > buffers, with an additional 1K words wired down for the graphics > buffer. It is fairly complete as old-school Forths go, but I didn't > bother to implement CREATE .. DOES> . The thread format is native > machine instructions - calls to primitives are JMP > > instructions, colon defs are called by CAL , constants and literals > are LAC , and variables are LAW . To save space, it uses the ancient > 3-characters+count header format, with each header taking exactly 2 > words. > > FIND is 22 instructions. NEXT is 1 instruction in the usual case, and > 3 for constants and variables. Pushing onto the stack is cheap, but > popping takes 6 instructions due to the way that auto-incrementing > (but not auto-decrementing) index registers work. BRANCH and ?BRANCH > are pretty conventional for a threaded code Forth - the branch offset > follows the primitive in the body. > > The turtle graphics package is 113 lines of Forth code, compiling to > 700 words. > > You can load files from emulated paper tape. You can run conditionals > and loops directly from interpret state using the temporary > compilation technique that I developed for Open Firmware. Here is a > Turtle Graphics drawing and its code: > > |: SQ 4 0 DO 200 FD 90 RT LOOP ; OK 8 0 DO SQ 45 RT LOOP OK | > Image: image (view on web) > <https://github.com/user-attachments/assets/55b55bce-5294-4b59-a55b-c591a5e0e3a3> > > > — > Reply to this email directly, view it on GitHub > <#206?email_source=notifications&email_token=ADUZY7BLZOUHSQMWJJVAUDL5RBYJFA5CNFSNUABBM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63RPGEYDQOJWG4YDHJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KYZTPN52GK4S7MNWGSY3L>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/ADUZY7FJYFZVUDFZ5RL2G5T5RBYJFAVCNFSNUABHKJSXA33TNF2G64TZHMZTCNBRGQ2DCNR3IRUXGY3VONZWS33OHMYTAOBZGY3TAM5BOYBA>. > Triage notifications, keep track of coding agent tasks and review pull > requests on the go with GitHub Mobile for iOS > <https://github.com/notifications/mobile/ios/ADUZY7GREQKHEMAV27UHUMD5RBYJFA5CNFSNUABBM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63RPGEYDQOJWG4YDHJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KUZTPN52GK4S7NFXXG> > and Android > <https://github.com/notifications/mobile/android/ADUZY7DT44KJ3JCB4LEGU7L5RBYJFA5CNFSNUABBM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63RPGEYDQOJWG4YDHJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2K4ZTPN52GK4S7MFXGI4TPNFSA>. > Download it today! > You are receiving this because you are subscribed to this > thread.Message ID: ***@***.***> > —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: ***@***.***> — Reply to this email directly, view it on GitHub <#206?email_source=notifications&email_token=ADUZY7DGHFQ5K3B4QQ5C25L5RHBC3A5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCOBWGMZDANZYUZZGKYLTN5XKOY3PNVWWK3TUUVSXMZLOOSWGM33PORSXEX3DNRUWG2Y#discussioncomment-18632078>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADUZY7BLDT4SFPTS6CSBQ3T5RHBC3AVCNFSNUABHKJSXA33TNF2G64TZHMZTCNBRGQ2DCNR3IRUXGY3VONZWS33OHMYTAOBZGY3TAM5BOYBA>. Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS <https://github.com/notifications/mobile/ios/ADUZY7ATWJ7E2JT4L7TXRVD5RHBC3A5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCOBWGMZDANZYUZZGKYLTN5XKOY3PNVWWK3TUUVSXMZLOOSVGM33PORSXEX3JN5ZQ> and Android <https://github.com/notifications/mobile/android/ADUZY7BUZSOWOHHYPMJLMT35RHBC3A5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCOBWGMZDANZYUZZGKYLTN5XKOY3PNVWWK3TUUVSXMZLOOSXGM33PORSXEX3BNZSHE33JMQ>. Download it today! You are receiving this because you commented.Message ID: ***@***.***>

[larsbrinkhoff]

I wrote a Forth for the IMLAC PDS-1 a while back. That machine is much like a 16-bit version of a PDP-7, with an added display processor, CRT for vector graphics, and keyboard. I elected to use subroutine threading with inlining of some primitives. I used an auto incrementing index location for the stack pointer, so e.g. + becomes the single instruction ADD I SP, or sp ) add, in my postifx cross assembler. Having done all this and happily coding away, I realised this boxes me into putting code in only the first 2K of memory. I wanted to add turtle graphics and have this run on the machine at the Interim Computer Museum (https://icm.museum/), but now I'm not so sure.

[larsbrinkhoff]

I chose to use the native JMS instruction for subroutine calls. Like on the PDP-7, the return address is stored in the first word of the subroutine; it can be regarded as the return stack TOS. Perhaps surprisingly, only a few words like EXECUTE have to be made reentrant by pushing this word to the return stack. Of course, this puts an end to many return stack tricks.