'Master' and 'slave': Tech terms face scrutiny amid anti-racism efforts

[PlainName]

It's never clear whether TX on a schematic is the *transmit out*,  or *transmit signal in*.

It should always be out on a DTE (terminal), in on a DCE (modem).

If you have a problem with that then you'll also likely have a problem with the master/slave style too because you need to know which is the master and which the slave to work out what's in or out.

[PlainName]

TX is always output, RX is always input. Using name "TX" as "transmit signal in" sounds like an error, unless it's just a signal between the transmitting device and the physical interface driver chip.

I'm going to be diplomatic here: wrong wrong wrong wrong wrong wrong. And it's wrong

[madires]

There are more problems with banning words. It's censorship! Those who want to rename an SPI master or a master branch because they think it might be racist are actually trying to enforce censorship which makes it even worse.

[Siwastaja]

It's never clear whether TX on a schematic is the *transmit out*,  or *transmit signal in*.

It should always be out on a DTE (terminal), in on a DCE (modem).

This is a source of confusion.

DTE/DCE classification is a higher level standard introduced by IBM, between computers or terminals, and modems, implementing kind of "master/slave" roles I specified above. Note, flow control is added as well. They arbitrarily decided to connect "TX" to "TX", and "RX" to "RX", so that you have to know the role (DTE vs. DCE) of the devices; only after that you can "simply" connect wires with same names together; and connecting DTE to DTE requires a special "crossover cable".

But all of this is utterly irrelevant, and only serve to confuse, in most use cases where "UART" interfaces are met today, where they are just bidirectional data links, nothing less, nothing more, and it's completely up to the designer of the higher level what relationships to introduce, under what terms. Younger generation designing UART things all the time know do not even have a freaking idea what we are talking about. So, copying the DTE/DCE terminology is not a sane choice when, for example, communicating between two microcontrollers on a PCB. If you want to take the IBM terminology, I suggest you use the typical 9/25 pin D connector types as well, with stickers showing whether your devices are DTE or DCE. Quite a niche today, really.

In this sense, we are both correct and wrong; just talking about different layers. To further clarify, my initial comment is neither right or wrong; it's not an official standard, it's a suggestion and describes how I and many others name things, and I truly think it's for the least amount of confusion. (Always when in doubt, be explicit; hence "cpu_to_modem" for example!)

[madires]

When you want to connect two DTEs you simply use a null modem cable. BTW, the gender of the DB9/25 connector already indicates if it's a DTE or DCE.

[tom66]

It's never clear whether TX on a schematic is the *transmit out*,  or *transmit signal in*.

It should always be out on a DTE (terminal), in on a DCE (modem).

If you have a problem with that then you'll also likely have a problem with the master/slave style too because you need to know which is the master and which the slave to work out what's in or out.

Master and slave is different, because it's clear that a master always initiates the transmission and the slave never does so.  It perfectly describes the arrangement.  An alternative description is "boss" and "worker" :-)

With TX and RX there have been ambiguous designs that I have reviewed by other engineers where the TX/RX on the connector is effectively reversed because it's matching the pinout of another connector.  Then having CPU_DATA_IN and CPU_DATA_OUT perhaps makes more sense but is a little wordy.  MISO and MOSI are considerably easier to understand.

[Siwastaja]

MISO and MOSI are great names because they explicitly spell out the whole thing. MI and MO, or SO or SI would have been enough; or even worse, many others would have taken an assumed viewpoint (master, for example), like IBM did, allowing the wires to be just I and O - but Motorola thought, what the heck, let's be super explicit about it. I like that.

If you think about it, RX and TX are names that cannot be properly used between devices; they describe the device interface.

The IBM DTE/DCE nomenclature swaps the logic of naming when it comes to DCE, so now the device interface is misnamed, but this makes using "RX" and "TX" as names on the wires between devices, possible (with an implicit viewpoint of DTE). IMHO, this was a stupid decision.

I have not seen this elsewhere; for example, all microcontroller UART interfaces specify TX and RX in a way so they are out and in, respectively. This way, you don't need to configure the interface to be "DTE" or "DCE", but you need to connect TX to RX, and RX to TX (which is, obviously, highly logical and causes less confusion than the IBM's "sometimes connect TX to TX and RX to RX and sometimes connect TX to RX and RX to TX" logic). Some modern UART peripherals do have a mode of swapping TX and RX, but that seems to be to address for design mistakes without requiring PCB respins; because of the confusion initiated by IBM. SPI interfaces do not have such swapping features because mistakes are very rare thanks to highly logical and explicit naming.

[vodka]

superficial differences such as skin colour don't matter

This is a textbook example of unconscious bias. Do you know why the police are more likely to stop and search (or kill in attempted self defense) certain ethnicities over others? Not because they are racist dickheads who read 4chan in their spare time, but precisely because some statistician somewhere had said that those races are more likely to commit crime. Or, dunno, ever heard of the concept of "white flight"? Pretty much what you have done, in different circumstances.



[/quote]

The policemen don't need to read 4chan for being informed , they are living themselves each day. And like said Lenin: the facts are stubborn .

Truly liberal countries like Sweden at least made a good call and banned collection of any racial statistics whatsoever. If race doesn't exist than the risk of carrying corona can't be broken down by race. I can appreciate the honesty and integrity in that.

  They are more zones NO-GO, the rape rate more higher of europe and the far-right is growing in each election. These symptoms are a country what don't work fine

[PlainName]

so that you have to know the role (DTE vs. DCE) of the devices; only after that you can "simply" connect wires with same names together; and connecting DTE to DTE requires a special "crossover cable"

Exactly the same with MOSI. You have to know if that bit of kit is a receiver or transmitter, master or slave, sender or consumer, and then you know which way a particular pin is going. Any confusion is merely that you don't know what kind of device you're connecting to.

Your beef is actually that the higher-level protocol labeling doesn't lend itself to describing the low-level pin characteristics. Well, duh. Use a different - low-level - label, is the answer. It's surely not too hard to put "DataOut" or "DataIn" on the chip pin and "Serial Tx" on the schematic net.

[PlainName]

MISO and MOSI are great names because they explicitly spell out the whole thing.

You need to know whether a device is a master or slave, no? And when a microcontroller can be a master and slave, what is the pin function with the label MISO?

[Siwastaja]

so that you have to know the role (DTE vs. DCE) of the devices; only after that you can "simply" connect wires with same names together; and connecting DTE to DTE requires a special "crossover cable"

Exactly the same with MOSI. You have to know if that bit of kit is a receiver or transmitter, master or slave, sender or consumer

No, no! MOSI always connects to MOSI, MISO always connects to MISO; the names are the same on typical master devices (like microcontrollers), and slave devices (like sensors). That's the whole point. No crossover connections ever needed.

That shifts the responsibility elsewhere, though: to the chip interface designer. They need to know that if they are designing a master, MISO will be an input, but if they are designing a slave, MISO will be an output.

You need to know whether a device is a master or slave, no? And when a microcontroller can be a master and slave, what is the pin function with the label MISO?

If the microcontroller is master, MISO is input: "master in"; if the microcontroller is slave, MISO is output ("slave out"). The great thing is the specification is fully supplied in the name itself, so it's very easy to remember. That's exactly the point about "explicit". The IBM TX/RX naming carries a hidden viewpoint which isn't in the name.

[PlainName]

If the microcontroller is master, MISO is input: "master in"; if the microcontroller is slave, MISO is output ("slave out"). The great thing is the specification is fully supplied in the name itself

I will hold my hands up to saying that completely passed me by, presumably from not actually using a masmicro as a slave - the MO has always been out.

[Siwastaja]

I regularly use microcontrollers in both master and slave configurations for SPI.

Although master obviously is a more common configuration because that's what you use when interfacing with most existing chips, like sensors, displays, etc.

SPI is great for transferring memory structures between two microcontrollers, usually with DMA, so that both parties can work "almost" like it was a single-MCU solution. In this case, one of the MCUs is obviously a slave, but both equally share their memories, the only difference is that master is the one driving the clock and chip select. (STM32 chips, for example, add a support for hardware checksum generation and checking, to make such use case more robust.)

[magic]

Perhaps the best systems just provide a framework to let people do their own thing...

At some point those systems inevitably start to believe that radicals are people too and then the fun begins

[retrolefty]

 My memory of the RS232 standard ( DTR, RTS, CTS, etc) was an ATT definition dating from the 1960s, IBM had nothing to do with it. In those days one classified a specific device as a terminal device or a communication device, where a 'computer' was not specifically defined as either one, as a computer could in practice be hardwired to terminals or modems or both.

 As a field engineer in the 70s I was often required to wire serial links between different manufactures devices and it was almost always a hassle without schematics to determine if a given device's connector pin wired to a physical electrical input or output. Many times the control pins were not even used by a device so things like flow control didn't even function and if needed would depend on software flow control using Xon/Xoff ASCII characters if supported by both sides on the link. Also connector gender did not define it's classification as male to male or female to female cables would often have to be made.

 Bottom line is the setting up RS232 links was mostly a snake pit/ time sink that the SPI interface naming solved.
 nicely. 

[Siwastaja]

My memory of the RS232 standard ( DTR, RTS, CTS, etc) was an ATT definition dating from the 1960s, IBM had nothing to do with it.

DCE/DTE classification and related naming in wiring has nothing to do with RS232, which is just an electrical signalling specification, though. See https://en.wikipedia.org/wiki/Data_terminal_equipment for more information, there is no source cited on that IBM claim though. D25 or D9 connector based DCE/DTE was a "complete" system so that the connector gender specified which one is DTE, which one DCE.

The complexity of such historical reliefs of course only shows how irrelevant the internal naming of such IBM (or otherwise) system when deciding how to label things in modern designs containing UART interfaces.

[coppice]

My memory of the RS232 standard ( DTR, RTS, CTS, etc) was an ATT definition dating from the 1960s, IBM had nothing to do with it.

DCE/DTE classification and related naming in wiring has nothing to do with RS232, which is just an electrical signalling specification, though. See https://en.wikipedia.org/wiki/Data_terminal_equipment for more information, there is no source cited on that IBM claim though. D25 or D9 connector based DCE/DTE was a "complete" system so that the connector gender specified which one is DTE, which one DCE.

The complexity of such historical reliefs of course only shows how irrelevant the internal naming of such IBM (or otherwise) system when deciding how to label things in modern designs containing UART interfaces.

RS-232 defines DTE, and DCE, electrical signalling (both async and sync), the use of the 25 pin D type connector and its pin out, the various signals like RTS and CTS, all in the context of a modem communicating with a terminal, and needing to signal things like ring, carrier and flow control.

[Siwastaja]

RS-232 defines DTE, and DCE, electrical signalling (both async and sync), the use of the 25 pin D type connector and its pin out, the various signals like RTS and CTS, all in the context of a modem communicating with a terminal, and needing to signal things like ring, carrier and flow control.

Oh, indeed you are right, thanks.

[julianhigginson]

Adafruit are replacing Master/Slave and other terms in all their documentation:
https://www.adafruit.com/blacklivesmatter

Cool! I guess now we get to see if any of the ridiculous slippery slopes that certain conservatives have been yelling about this topic actually come true....  My guess is, they won't.

I'm still unsure of how much of a push there is from actual people of colour for this change, and therefore how much a of a need for it there is - but since it's so easy to do, and has some precedent in the community now, anything I have a say in from now on will not use master/slave terminology.

I'm warming to controller/peripheral.
CIPO, PICO. done.

[Nominal Animal]

I read this today. It might have relevance to some interested in this thread. For those willing to engage in some self discovery there is a link in there to some tests you can do to reveal your own implicit biases. You will find you're not special after all. And I don't mean "special".

https://www.pbs.org/newshour/nation/making-people-aware-of-their-implicit-biases-doesnt-usually-change-minds-but-heres-what-does-work

Did you notice Anthony Greenwald, the leading implicit bias expert, says that the only long-term solution is essentially to ignore what groups the applicant belongs to (gender, race, culture), and instead only look at their output?  That he's "very sceptical" about "implicit bias training"?

That's exactly what I've been saying we should do: treat each others as individuals, evaluate others based on their actions and interactions with others, and keep stuff like gender, skin color, race, "oppressed group" as not-work-relevant personal attributes only.  You know, equality of opportunity, and to hell with equity/equality of outcome.

have even thought about running as an independent myself so I can vote for myself 

I'd hate for someone like me getting real political power; they'd be a straight up dick-tator. 

When there was talk about us becoming a republic, one of the names put forward for president was Dick Smith. He famously said he'd only do it if he could be a dictator, as it's the only way to get things done!

Exactly.  I can't help but try and fix problems, and I'm not above manipulating people if I think it is for their own good.

[EEVblog]

I'm still unsure of how much of a push there is from actual people of colour for this change, and therefore how much a of a need for it there is - but since it's so easy to do, and has some precedent in the community now

It's not "easy to do". There are generations of course material, datasheets, app notes, and tutorials etc, not to mention the countless teachers and engineers who don't give a rats arse about modern identity politics. Adafruit and a few others wouldn't even represent a fraction of a percent of the material and sentiment out there.
And not teaching the accepted industry standard terms is disadvantageous to any students.

[james_s]

Yes confusion is a problem, and yes once half the language has been banned everything will just be called "thing", then we will all know exactly what we are talking about.

[julianhigginson]

It's not "easy to do". There are generations of course material, datasheets, app notes, and tutorials etc, not to mention the countless teachers and engineers who don't give a rats arse about modern identity politics. Adafruit and a few others wouldn't even represent a fraction of a percent of the material and sentiment out there.
And not teaching the accepted industry standard terms is disadvantageous to any students.

it *is* easy to do going forward.

I'm not sure how many people are talking about going back and pulping every physical or digital copy of every technical book or manual that uses the old terms... Or arresting people who use the old terms... that would be ridiculous, but I've only see that slipperly sloped, not pushed by anyone who is in favour or even ambivalent about it.

If I write a document talking about SPI controller and peripheral, and someone doesn't understand what I'm talking about then I've got very grave concerns for what happens when they come across an SPI device that has weird timing requirements or unusual data frame lengths and enable pin activity requirements that don't match both ends of the connection... letalone understanding the 4 basic clock/data options that are used everywhere.

And the good news is, once the new terms get decided on (people will come to a consensus about what they all want to use) it'll become pretty easy to tell what most people are doing, whether they describe it in old words or new words.

It's language. It's always changing. It changes for plenty of reasons. Getting upset about *one* instance of words changing for the specific purpose of relieving some perceived level of suffering in some group of people seems (even if it seems like "tokenism" or 'white saviourship virtue signalling" does that even matter that much?) seems, well, a little bit off.  because no matter how ridiculous it might be, it's really not going to cause anyone reading this any measurable harm.

[james_s]

It's not "easy to do". There are generations of course material, datasheets, app notes, and tutorials etc, not to mention the countless teachers and engineers who don't give a rats arse about modern identity politics. Adafruit and a few others wouldn't even represent a fraction of a percent of the material and sentiment out there.
And not teaching the accepted industry standard terms is disadvantageous to any students.

Also it adds up. There's this little change and that little change and some other little change, it's like saying "well this only costs $5 a month!" Ok yeah, $5 here, $10 there, it adds up and pretty soon it's not such a trivial amount of money. If it takes me cumulatively a couple of hours to adjust to a new set of words for something which I don't think is unreasonable when you consider dealing with all these "microtranslations" each time I encounter the old term that's a couple of hours out of my finite life that I could have spent learning something I didn't already know. And why? Because someone found the old words offensive? What if I find the new words offensive? I do in fact find change that is just for the sake of change offensive. Are my feelings less important than those of someone else?

[julianhigginson]

https://yourlogicalfallacyis.com/slippery-slope