My favorite Tcl story is a little side note about how Tcl could have been the language of the web[1]
the founding of Netscape occurred at the same time I was deciding where to go in industry when I left Berkeley in 1994. Jim Clarke and Marc Andreessen approached me about the possibility of my joining Netscape as a founder, but I eventually decided against it (they hadn't yet decided to do Web stuff when I talked with them). This is one of the biggest "what if" moments of my career. If I had gone to Netscape, I think there's a good chance that Tcl would have become the browser language instead of JavaScript and the world would be a different place! However, in retrospect I'm not sure that Tcl would actually be a better language for the Web than JavaScript, so maybe the right thing happened.
Too humble Dr. Ousterhout! It would have been a far better language.
I was leading a Tcl project around then, and, though there were some very neat things about the unusual Tcl evaluation model, I wasn't a fan of using it for nontrival work. For example, it wasn't a natural fit for working with graph structures like I had to, and like you might want for browser DOM.
(That said, Tcl would've been much better than JS, and I suspect that Ousterhout would've figured out some smart things to make it good for the browser.)
Maybe 5 years later, I was meeting with Tim-Berners Lee, and I kinda pitched Scheme to him, without planning to, but he was very interested when he asked what I'd been working on.
But then he went and did a conference keynote, in which he promoted Python as the language for ordinary people doing Web stuff. And I think he referenced one of the things I'd written in support of Scheme... as an anti-requirement for his populist vision for the Web. :)
(I wish I could've been involved in that, because I could've made a case for a populist spin on Scheme at the time.)
Funny that Tcl was mentioned today. I've recently been hacking on jimtcl[1] (a small footprint implementation of Tcl) to make objects multi-threaded safe, as that's the core language that folk.computer[2] uses (the main authors are making the interpreter and reactive DB parallelizable rn, so I've been tinkering with how to reduce copying).
I have known about Tcl since the 1990s. I remember Stallman advising not use it. Maybe he did not like the license?
Yesterday, over 30 years later I was forced to write my first code. Noticed that a Cisco switch had an incredibly bad ssh implemtation, so to automate some commands I needed expect scripting. Expect is based on Tcl.
If I recall correctly, I think he was just more of a lisp fan and promoted something like Guile or E-Lisp. It may have been license as well. I'm not very sure about it though.
I think he was worried about Sun Microsystems involvement[0] w Tcl when John Ousterhout went there[1], too. But mostly seems like a collection of unsolicited strawmen[2]. Also probably love-of-lisp.
I own the first edition of this book and it's really good. I've written a few tcl scripts with it, but nothing major. Python just has the better ecosystem now for my needs. I might use tcl if I needed something really small though.
It is still actively developed and used, yeah. People tend to mention the fpga scene most, tcl is the predominant language there, but I use it mostly with or along side audio/visual software.
I would be interested to know if anyone ever built anything non-trivial that didn't turn into a complete mess due to TCL's general type-flimsyness and "everything is a string" philosophy.
The GUI of Pure Data (https://en.m.wikipedia.org/wiki/Pure_Data) is written in Tcl/Tk. In the 90s this was a popular and sensible choice, but not so much in 2025 :) I still have hopes that one day we'll manage to replace it with something better.
(There are alternative implementations of Pd, such as PurrData or PlugData, that use different UI frameworks.)
TCL/tk is used for the vast majority of integrated circuit verification tools. This is used by all the large manufacturers, even competitors use each others tools.
Absolutely huge codebases, there's git blames that go back to the 80s.
Someone did a nice job of porting your revision control data. It would be fascinating to know what the previous systems were - probably svn or cvs, maybe both in chronological order, but before that?
There is no big difference between TCL and other languages like Python in this respect. Yes, TCL stores everything as a string, but it also assigns a type when the data is used. If you need a defined type, you can use structures; there are even classes/objects available as packages.
Yep! It's called shimmering. The way it works is it has a string and internal representation. When the internal representation is modified it invalidates the string version. If the string version is needed later it'll regenerate the string representation. That way it can always be used as a string, but internally it is pretty efficient when it stays as the same type.
You’re misunderstanding. Dynamic languages inherently have a single type that is checked _at runtime_. Store it in a string, store it in a struct. It doesn’t matter.
For FPGA IDEs, a lot of the background is just tcl scripts so you can easily write your own build scripts. You could theoretically put them into your CI pipeline but I don't know if anyone actually does this.
TCL stands for "tool command language." In terms of non trivial tools that have been built with it there have been several. AOLServer, MacPorts, tons of CAD and engineering software. There's a whole wikipedia category just for this:
Aside from that there was Tk which was used to create many UIs in an era where scripted UIs were otherwise painful to create and maintain. A prominent example here is 'xconfig' from the linux kernel.
Like any good environment a little bit of practice with it can see you overcome it's apparent shortcomings and see you producing reliable and useful software in far less time that you might have in a more traditional way.
Programming is about trade offs. It's not a checklist of "good ideas."
My big project at work uses pytk only because it was already included with the anaconda distribution. It works fine, it's not pretty but it's just an internal engineering tool.
My favorite Tcl story is a little side note about how Tcl could have been the language of the web[1]
Too humble Dr. Ousterhout! It would have been a far better language.1 - https://pldb.io/blog/JohnOusterhout.html
I was leading a Tcl project around then, and, though there were some very neat things about the unusual Tcl evaluation model, I wasn't a fan of using it for nontrival work. For example, it wasn't a natural fit for working with graph structures like I had to, and like you might want for browser DOM.
(That said, Tcl would've been much better than JS, and I suspect that Ousterhout would've figured out some smart things to make it good for the browser.)
Maybe 5 years later, I was meeting with Tim-Berners Lee, and I kinda pitched Scheme to him, without planning to, but he was very interested when he asked what I'd been working on.
But then he went and did a conference keynote, in which he promoted Python as the language for ordinary people doing Web stuff. And I think he referenced one of the things I'd written in support of Scheme... as an anti-requirement for his populist vision for the Web. :)
(I wish I could've been involved in that, because I could've made a case for a populist spin on Scheme at the time.)
In that world, Netscape might have also acquired Naviserver instead of AOL. Wasn't the plan at Netscape to make money on their server software?
Funny that Tcl was mentioned today. I've recently been hacking on jimtcl[1] (a small footprint implementation of Tcl) to make objects multi-threaded safe, as that's the core language that folk.computer[2] uses (the main authors are making the interpreter and reactive DB parallelizable rn, so I've been tinkering with how to reduce copying).
[1] https://github.com/msteveb/jimtcl [2] https://folk.computer/
I have known about Tcl since the 1990s. I remember Stallman advising not use it. Maybe he did not like the license?
Yesterday, over 30 years later I was forced to write my first code. Noticed that a Cisco switch had an incredibly bad ssh implemtation, so to automate some commands I needed expect scripting. Expect is based on Tcl.
Most IOS implementations (classic, XE, XR) also have tclsh, a tcl REPL (that can also interact with the file systems to load and write scripts).
If I recall correctly, I think he was just more of a lisp fan and promoted something like Guile or E-Lisp. It may have been license as well. I'm not very sure about it though.
I think he was worried about Sun Microsystems involvement[0] w Tcl when John Ousterhout went there[1], too. But mostly seems like a collection of unsolicited strawmen[2]. Also probably love-of-lisp.
[0] That any proximity to commerce is evil?
[1] https://en.wikipedia.org/wiki/John_Ousterhout
[2] https://vanderburg.org/old_pages/Tcl/war/
I own the first edition of this book and it's really good. I've written a few tcl scripts with it, but nothing major. Python just has the better ecosystem now for my needs. I might use tcl if I needed something really small though.
Is Tcl still a thing? At my first job I extended with a set of data-processing commands.
It is still actively developed and used, yeah. People tend to mention the fpga scene most, tcl is the predominant language there, but I use it mostly with or along side audio/visual software.
I had this book, it's well written.
> this immensely flexible and versatile language
I would be interested to know if anyone ever built anything non-trivial that didn't turn into a complete mess due to TCL's general type-flimsyness and "everything is a string" philosophy.
Cisco iOS[0], F5 iRules[1], Tealeaf network monitoring/processing[2][3], Fidessa[4][5], …
[0] https://www.cisco.com/c/en/us/td/docs/ios-xml/ios/ios_tcl/co...
[1] https://community.f5.com/kb/technicalarticles/irules-concept...
[2] https://www.acoustic.com/tealeaf
[3] https://help.goacoustic.com/hc/en-us/articles/360043279654-C...
[4] https://www.cmegroup.com/solutions/market-tech-and-data-serv...
[5] http://www.services.fidessa.com/brochures/_training/VF043.pd...
The GUI of Pure Data (https://en.m.wikipedia.org/wiki/Pure_Data) is written in Tcl/Tk. In the 90s this was a popular and sensible choice, but not so much in 2025 :) I still have hopes that one day we'll manage to replace it with something better.
(There are alternative implementations of Pd, such as PurrData or PlugData, that use different UI frameworks.)
git-gui and gitk are also written in Tcl/Tk.
TCL/tk is used for the vast majority of integrated circuit verification tools. This is used by all the large manufacturers, even competitors use each others tools.
Absolutely huge codebases, there's git blames that go back to the 80s.
Someone did a nice job of porting your revision control data. It would be fascinating to know what the previous systems were - probably svn or cvs, maybe both in chronological order, but before that?
Flightaware, at one point, had a lot of Tcl code
https://www.flightaware.com/about/code/
There is no big difference between TCL and other languages like Python in this respect. Yes, TCL stores everything as a string, but it also assigns a type when the data is used. If you need a defined type, you can use structures; there are even classes/objects available as packages.
Does it still do this? I thought they changed this many years ago?
Yep! It's called shimmering. The way it works is it has a string and internal representation. When the internal representation is modified it invalidates the string version. If the string version is needed later it'll regenerate the string representation. That way it can always be used as a string, but internally it is pretty efficient when it stays as the same type.
That is absolutely not how Python works. Python values are typed.
> "a dynamically typed language is a statically typed language with only one static type."
Yes but we are not talking about static types...
You’re misunderstanding. Dynamic languages inherently have a single type that is checked _at runtime_. Store it in a string, store it in a struct. It doesn’t matter.
The runtime values have tags that denote their runtime type. They are not of a single type. You are talking about static types.
For FPGA IDEs, a lot of the background is just tcl scripts so you can easily write your own build scripts. You could theoretically put them into your CI pipeline but I don't know if anyone actually does this.
Sqlite started as a tcl extension. That isn't exactly your question, but I thought it was closely related.
Wavesurfer was a great program written Tcl/Tk, with an audio processing library called "snack" in C with Tcl wrappers.
https://en.wikipedia.org/wiki/WaveSurfer
A lot of old GUIs are built on Tcl/Tk, lots of people use matplotlib which I think uses it by default.
TCL stands for "tool command language." In terms of non trivial tools that have been built with it there have been several. AOLServer, MacPorts, tons of CAD and engineering software. There's a whole wikipedia category just for this:
https://en.wikipedia.org/wiki/Category:Free_software_program...
Aside from that there was Tk which was used to create many UIs in an era where scripted UIs were otherwise painful to create and maintain. A prominent example here is 'xconfig' from the linux kernel.
Like any good environment a little bit of practice with it can see you overcome it's apparent shortcomings and see you producing reliable and useful software in far less time that you might have in a more traditional way.
Programming is about trade offs. It's not a checklist of "good ideas."
My big project at work uses pytk only because it was already included with the anaconda distribution. It works fine, it's not pretty but it's just an internal engineering tool.
For many years every web page at CNN and all web email at AOL was hosted by AOLserver and coded in TCL. It was not a mess.
what is your definition of non trivial?