Jun 14, 2026 · 25 min · 5 segments
We get into the process of working on and with formal protocol specifications, something Andy is familiar with from his work on Matrix. Support us on Patreon and get an ad-free…
So I guess we should start off by talking about what kind of spec we're talking about.
Not spec-driven development.
[chuckles] Not spec-driven development, but also whenever you're doing any software project, you need to have some kind of idea what you're doing, and people often talk about that as a spec, and maybe that is part of the conversation we're having here.
But my understanding of your question, Amolith, is you're talking about something that is going to become part of, like, a formal specification that's published that other people are gonna implement too, not just your own code.
Is that right?
Yes, more like a protocol specification rather than an implementation specification.
So more standards rather than-
Mm-hmm
...
this is what we're gonna do for our particular project.
Right.
So yeah, in the Matrix world, there's a, a way of approaching this that has been become established as the norm, and I think it's partly based on, like, bad experiences with other spec processes in the past.
There's sort of some people involved with Matrix who've been involved with things like the IETF and things like that previously.
N- not to say that that's, those processes are bad, but they can sometimes be quite painful and slow and even sometimes diverge from what's actually getting implemented.
So in the Matrix world, you can write a piece of spec whenever you like, a proposal for a change to the spec.
But if it's gonna get merged as a formal part of the spec, there needs to be an implementation demonstrating that it makes sense before you can do it.
And I've actually seen a similar movement in the C++ community where some features were introduced without anyone actually implementing them in compilers, and then at some point they realized they couldn't implement it in the compiler, and it never got done.
So yeah, I think there's a movement towards, yeah, we need to be a bit more careful to make sure things are realistic and, and fully thought through, which you can only really do by implementing the thing at some point in the process.
Yeah, Amolith, you originally asked something along the lines of, do you write the spec first and then implement, or do you implement first and write the spec to that? And at least in the little bits that I've worked, and it hasn't been, like, formal specifications or anything, it's been a little bit of both.
Uh, for exactly the reasons that Andy just stated, you might have an idea of some specs, you might write a little bit of it, and then you have to implement some of that to make sure it's realistic.
And you might come up with eventually what is a, like, good base spec and a good base, like, prototype implementation that you can then go and, like, massage either or.
At least the projects that I've worked on before, they have then tried to polish off the spec first and then redo the implementation based off the spec.
And that sort of, like, cyclical approach to it has been the best results that I've seen.
If you try to do just the spec by itself, you will inevitably run into areas where when you try to implement it, it's just unrealistic.
It, it doesn't work that way.
So I guess we should start off by talking about what kind of spec we're talking about.
Not spec-driven development.
[chuckles] Not spec-driven development, but also whenever you're doing any software project, you need to have some kind of idea what you're doing, and people often talk about that as a spec, and maybe that is part of the conversation we're having here.
But my understanding of your question, Amolith, is you're talking about something that is going to become part of, like, a formal specification that's published that other people are gonna implement too, not just your own code.
Is that right?
Yes, more like a protocol specification rather than an implementation specification.
So more standards rather than-
Mm-hmm
...
this is what we're gonna do for our particular project.
Right.
So yeah, in the Matrix world, there's a, a way of approaching this that has been become established as the norm, and I think it's partly based on, like, bad experiences with other spec processes in the past.
There's sort of some people involved with Matrix who've been involved with things like the IETF and things like that previously.
N- not to say that that's, those processes are bad, but they can sometimes be quite painful and slow and even sometimes diverge from what's actually getting implemented.
So in the Matrix world, you can write a piece of spec whenever you like, a proposal for a change to the spec.
But if it's gonna get merged as a formal part of the spec, there needs to be an implementation demonstrating that it makes sense before you can do it.
And I've actually seen a similar movement in the C++ community where some features were introduced without anyone actually implementing them in compilers, and then at some point they realized they couldn't implement it in the compiler, and it never got done.
So yeah, I think there's a movement towards, yeah, we need to be a bit more careful to make sure things are realistic and, and fully thought through, which you can only really do by implementing the thing at some point in the process.
Yeah, Amolith, you originally asked something along the lines of, do you write the spec first and then implement, or do you implement first and write the spec to that? And at least in the little bits that I've worked, and it hasn't been, like, formal specifications or anything, it's been a little bit of both.
Uh, for exactly the reasons that Andy just stated, you might have an idea of some specs, you might write a little bit of it, and then you have to implement some of that to make sure it's realistic.
And you might come up with eventually what is a, like, good base spec and a good base, like, prototype implementation that you can then go and, like, massage either or.
At least the projects that I've worked on before, they have then tried to polish off the spec first and then redo the implementation based off the spec.
And that sort of, like, cyclical approach to it has been the best results that I've seen.
If you try to do just the spec by itself, you will inevitably run into areas where when you try to implement it, it's just unrealistic.
It, it doesn't work that way.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.