Legacy Code Documentation
-
Questions Is it Maintained? Do you look for it or just explore the code? What do you do if it is not there? Is the code documenting itself good enough?
Most developer document the wrong things. Far too often, they repeat what the code line says, whether like "Here we set i to the value 5" or repeat the parameters of the method heading, "explaining" e.g. "numberOfApples: The number of apples". Or, when asked for more high level, code independent documentation, they present a block chart with Application, Database and Network modules. What other developers need lies in between: A documentation of the purpose of each code module, of its interaction with other modules. Replace 'module' with 'class' as appropriate. The data structures - how they are interlocked, what they represent as a whole - that usually means, across several object classes. To draw a line: I want off line - in the sense: off the code lines - documentation to go down to the deepest level where it is still independent of the programming language. You should be able to build those modules, those data structures, in Python, in C++, even in Visual Basic. You document the interfaces between modules, not only the syntax (name, parameters) but the dynamics as well. That will help the next code maintainer on the right track! And, these interactions are far more stable than the individual code lines; you can make a lot of changes without invalidating the documentation. Every now and then it may need to be extended with new interactions or parameter options, yet all the old stuff remains valid. When you get down to code, you may of course add code comments - but without repeating the off code line documentation; there is no need to. Obviously, you should avoid 'Here we set i to 5' style comments, and you should use descriptive type and variable names. When you delete code lines, you delete the accompanying comments. When you change code lines, you change the accompanying comments. End-of-line comments (lined up at col. 70 or 80 not to blur the code) is a lot better for keeping them up to date than pushing them to the top of the method. But these code comments are far less essential. You read them after you have studied all the off line documentation, and know the interactions, the data structures, the interfaces. What is left is to understand how these interfaces are realized, within a single module/class. That is usually a far simpler task than grokking the structure of the entire program system as a whole. I do like to have (i.e. both make and read) code comments, but if I could have a choice between code comments and a system documentation from immediately
-
"Is the code documenting itself good enough?" You're joking, right? I am the sole proprietor of a legacy system written in the 80s. Technically, I've retired, but I've given this one customer the option of lifting the production (note: production) system to today. It's a re-write (the core functions) but most of the code goes away due to a functional database, report writer and C++ or .net. The group involved is part of a huge international corporation, so crickets.
Charlie Gilley “They who can give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety.” BF, 1759 Has never been more appropriate.
-
MarcusCole6833 wrote:
Is the code documenting itself good enough?
I am personally a big fan of using descriptive variables such that the code becomes self-explanatory. I also have a standard set of names for common-use objects that could incriminate identify me as the author. Edit: That said, when the intent is not clear, or you/they are doing something 'clever', then documenting/commenting is encouraged. Comments may or may not get updated, but the code doesn't lie. :laugh:
"Go forth into the source" - Neal Morse "Hope is contagious"
kmoorevs wrote:
you/they are doing something 'clever', then documenting/commenting is encouraged.
I maintain as a strong rule: 'Clever' code should be avoided, if at all possible. If it can be rewritten in a more readable way, do so -- even if timing of that specific loop shows that it takes 40% more time. Most likely is is 0.01% of the total run time for the application.
Comments may or may not get updated, but the code doesn't lie. :laugh:
Well ... The comment may still be the honest truth of what the code was intended to do. The code may be honest about the coder's misconceptions, or possibly his intent to impress with his 'smartness', without succeeding :-)
Religious freedom is the freedom to say that two plus two make five.
-
MarcusCole6833 wrote:
Is the code documenting itself good enough?
I am personally a big fan of using descriptive variables such that the code becomes self-explanatory. I also have a standard set of names for common-use objects that could incriminate identify me as the author. Edit: That said, when the intent is not clear, or you/they are doing something 'clever', then documenting/commenting is encouraged. Comments may or may not get updated, but the code doesn't lie. :laugh:
"Go forth into the source" - Neal Morse "Hope is contagious"
kmoorevs wrote:
Comments may or may not get updated, but the code doesn't lie. :laugh:
Unless that code is being refactored and the naming butchered by some other dev. While I do agree good naming is important, it doesn't remove the need for comments completely. Blanket statements is typically the tell tale sign of the inexperienced.
Jeremy Falcon
-
Most developer document the wrong things. Far too often, they repeat what the code line says, whether like "Here we set i to the value 5" or repeat the parameters of the method heading, "explaining" e.g. "numberOfApples: The number of apples". Or, when asked for more high level, code independent documentation, they present a block chart with Application, Database and Network modules. What other developers need lies in between: A documentation of the purpose of each code module, of its interaction with other modules. Replace 'module' with 'class' as appropriate. The data structures - how they are interlocked, what they represent as a whole - that usually means, across several object classes. To draw a line: I want off line - in the sense: off the code lines - documentation to go down to the deepest level where it is still independent of the programming language. You should be able to build those modules, those data structures, in Python, in C++, even in Visual Basic. You document the interfaces between modules, not only the syntax (name, parameters) but the dynamics as well. That will help the next code maintainer on the right track! And, these interactions are far more stable than the individual code lines; you can make a lot of changes without invalidating the documentation. Every now and then it may need to be extended with new interactions or parameter options, yet all the old stuff remains valid. When you get down to code, you may of course add code comments - but without repeating the off code line documentation; there is no need to. Obviously, you should avoid 'Here we set i to 5' style comments, and you should use descriptive type and variable names. When you delete code lines, you delete the accompanying comments. When you change code lines, you change the accompanying comments. End-of-line comments (lined up at col. 70 or 80 not to blur the code) is a lot better for keeping them up to date than pushing them to the top of the method. But these code comments are far less essential. You read them after you have studied all the off line documentation, and know the interactions, the data structures, the interfaces. What is left is to understand how these interfaces are realized, within a single module/class. That is usually a far simpler task than grokking the structure of the entire program system as a whole. I do like to have (i.e. both make and read) code comments, but if I could have a choice between code comments and a system documentation from immediately
that's pretty religious ;) Seems to me a bit of a pet peeve of yours. Mine as well. I'll swing this in a different direction. There are a few people I've worked with that I term "whiz kids." Their minds work so fast, 5x faster than mine, that everything is intuitively obvious. Not one damn line of comments. No doc to save anyone's lives, nothing. I simply gave up. I completely concur about not documenting loops and i++, but module level comments are necessary. And ffs, tie the code comments back to external documentation. Now, 40 years later, no one gives an elephant. I've worked at (counting....) 9 companies and myself, and there is only one person I can count on to write proper documentation. It's simply pathetic.
Charlie Gilley “They who can give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety.” BF, 1759 Has never been more appropriate.
-
that's pretty religious ;) Seems to me a bit of a pet peeve of yours. Mine as well. I'll swing this in a different direction. There are a few people I've worked with that I term "whiz kids." Their minds work so fast, 5x faster than mine, that everything is intuitively obvious. Not one damn line of comments. No doc to save anyone's lives, nothing. I simply gave up. I completely concur about not documenting loops and i++, but module level comments are necessary. And ffs, tie the code comments back to external documentation. Now, 40 years later, no one gives an elephant. I've worked at (counting....) 9 companies and myself, and there is only one person I can count on to write proper documentation. It's simply pathetic.
Charlie Gilley “They who can give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety.” BF, 1759 Has never been more appropriate.
Contrary to my last(vastly misunderstood post), this is not apparent, it simply is (for real). I am an OF and have noticed how "newer generations" choose NOT to use (simple) old, proven methods , stuff. My manager did not use AI or RTFM - his "comment or else " was based on "what will happen if you get run over by a bus and your code is not documented / commented?" For those hard of hearing - undocumented / uncommented code was against company interest, period. I am not a wiz kid, but if the code task is to test HARDWARE , CLOSE TO REAL TIME, and if the code is not documented , commented AND is missing ANY time reference - it will not "just " APPARENTLY fail - it will fail for sure. IMHO self documented code is oxymoron
-
Never trust documentation. There is no guarantee the last developer updated it when they changed the code. You can read it as a guide. So trust, but verify.
Hogan
-
Questions Is it Maintained? Do you look for it or just explore the code? What do you do if it is not there? Is the code documenting itself good enough?
All documentation is out of date as soon as it's written. Code is never out of date.
-
Never trust documentation. There is no guarantee the last developer updated it when they changed the code. You can read it as a guide. So trust, but verify.
Hogan
I hate that, and it's why I lean toward not commenting, and expressing intent through code, wherever I can. Comments are a curse, even if they're sometimes necessary.
Check out my IoT graphics library here: https://honeythecodewitch.com/gfx And my IoT UI/User Experience library here: https://honeythecodewitch.com/uix
-
Questions Is it Maintained? Do you look for it or just explore the code? What do you do if it is not there? Is the code documenting itself good enough?
Is it really just the code, or a whole system? If a whole system, check and/or create a high-level overview, even if it "should be obvious" and make sure you have the documentation necessary to make a change, rebuild, test, distribute and install. (I.e. you have to go through that process to be sure it works). As for code, depends partly on how much expertise in the coding language(s) you think will be available if a change is necessary, particularly if people have used unusual language features or coded in an unusual way compared to "normal industry practice" - perhaps make a few general remarks noting such stuff.
-
Contrary to my last(vastly misunderstood post), this is not apparent, it simply is (for real). I am an OF and have noticed how "newer generations" choose NOT to use (simple) old, proven methods , stuff. My manager did not use AI or RTFM - his "comment or else " was based on "what will happen if you get run over by a bus and your code is not documented / commented?" For those hard of hearing - undocumented / uncommented code was against company interest, period. I am not a wiz kid, but if the code task is to test HARDWARE , CLOSE TO REAL TIME, and if the code is not documented , commented AND is missing ANY time reference - it will not "just " APPARENTLY fail - it will fail for sure. IMHO self documented code is oxymoron
What is an OF, and more to the point, what other post? Inquiring minds want to know (I looked at the thread, and this is your first post). But, I completely agree with your comment.
Charlie Gilley “They who can give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety.” BF, 1759 Has never been more appropriate.
-
Is it really just the code, or a whole system? If a whole system, check and/or create a high-level overview, even if it "should be obvious" and make sure you have the documentation necessary to make a change, rebuild, test, distribute and install. (I.e. you have to go through that process to be sure it works). As for code, depends partly on how much expertise in the coding language(s) you think will be available if a change is necessary, particularly if people have used unusual language features or coded in an unusual way compared to "normal industry practice" - perhaps make a few general remarks noting such stuff.
THIS!!! Code does not exist on it's own. It was written to solve a problem. If you don't tie the code back to the problem, you lose context. Sometimes this is called tribal knowledge. I'm watching an organization bleed to death, but the multinational corp above it is oblivious. As my boss told me, it's not up to you to help a multi-billion dollar company fix their stupidity. 2 weeks later, I jumped out of the plane (with a parachute) and retired. It's been two months now, and I have not received a single email. :)
Charlie Gilley “They who can give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety.” BF, 1759 Has never been more appropriate.
-
I hate that, and it's why I lean toward not commenting, and expressing intent through code, wherever I can. Comments are a curse, even if they're sometimes necessary.
Check out my IoT graphics library here: https://honeythecodewitch.com/gfx And my IoT UI/User Experience library here: https://honeythecodewitch.com/uix
There is a reasonable balance between documenting something and the overly detailed documentation that is likely to get out of date. I like the "here is a new thing and here is the general pattern of how it works" type of documentation. Code comments are reserved for "I know this is stupid, but users." And references to Kelly Clarkson to make the other devs laugh when they read my code. We have to have some fun around here.
Hogan
-
MarcusCole6833 wrote:
Is it Maintained?
Legacy code or not, if the code is updated the docs for it should be updated.
MarcusCole6833 wrote:
Do you look for it or just explore the code?
It should be a part of the code. jsdoc or doxygen style comments should be littered all throughout the code. There should also be READMEs in the project. No point in having to hunt down an external resource that would only get stale.
MarcusCole6833 wrote:
What do you do if it is not there?
Ask why it's not there with the original devs. If I'm now responsible for said legacy code and if there is some documentation somewhere then I'll try and add it to the project.
MarcusCole6833 wrote:
Is the code documenting itself good enough?
No. This goes back to the age old "what makes a good comment" issue. People that say code is always self-documenting are just wrong. They've never worked with other humans and certainly was never a professional dev. What makes a good comment? Not stuff like this:
// set x to 5
let x = 5;Well no duh. That is self-documenting and obvious, but stuff like this always isn't:
// keep the raw data for these, also bands should use the population
const average = [...simpleMovingAverage(sample, range)].pop()?.actual ?? 0 as Price;
const band = standardDeviation(sample.map((x: Market) => x.close), true).actual * deviations;Not everyone maintaining the code may know off the top of their head the note about how something should be implemented. One could argue RTFM, but let's be realistic... people maintaining code do not Google every last line of code. Also, there may be no well-known manual for some things. It's easy for someone to "fix" it and break it and even get past a PR. So, when you do something that may be considered unusual or haphazard or just only known by a select few having a comment should present to make it explicit. As such, I don't consider code always self-documenting in the real world.
Jeremy Falcon
Quote:
People that say code is always self-documenting are just wrong. They've never worked with other humans and certainly was never a professional dev.
Thank you, thank you, thank you! I've lost count of the number of arguments I've had about this. I've also lost count of the number times I've gone into my own code after some time and silently thanked Earlier Me for comments like
//Outputs have to match previous provider, so yes, converting this number to text here is correct
and
'NB loop starting at 1 is deliberate to skip headers
I used to often include links to Sharepoint documents in comments. Stopped doing that after a company "rationalised" their document storage (euphemism for moved to different provider and did not migrate all documents)
-
I hate that, and it's why I lean toward not commenting, and expressing intent through code, wherever I can. Comments are a curse, even if they're sometimes necessary.
Check out my IoT graphics library here: https://honeythecodewitch.com/gfx And my IoT UI/User Experience library here: https://honeythecodewitch.com/uix
In code reviews, I tell the developers to remove all comments except ones that explain _why_ we needed to do something. The _what_ should be evident from reading the code. The worse comments are ones that only re-phrase the name of the class or method. I deal with 30 year old code that is still being updated and maintained. You can tell lots of different programmers had their hands in the pot and not all of it is good code. For the first 20 years, no code reviews were done. I'm the last person on the team with more than 3 years experience (I have 24 years) on this codebase. If I leave, then they would be so lost in a lot of areas of the code. There is no way we could document all of it at this point.
Bond Keep all things as simple as possible, but no simpler. -said someone, somewhere
-
In code reviews, I tell the developers to remove all comments except ones that explain _why_ we needed to do something. The _what_ should be evident from reading the code. The worse comments are ones that only re-phrase the name of the class or method. I deal with 30 year old code that is still being updated and maintained. You can tell lots of different programmers had their hands in the pot and not all of it is good code. For the first 20 years, no code reviews were done. I'm the last person on the team with more than 3 years experience (I have 24 years) on this codebase. If I leave, then they would be so lost in a lot of areas of the code. There is no way we could document all of it at this point.
Bond Keep all things as simple as possible, but no simpler. -said someone, somewhere
-
Quote:
People that say code is always self-documenting are just wrong. They've never worked with other humans and certainly was never a professional dev.
Thank you, thank you, thank you! I've lost count of the number of arguments I've had about this. I've also lost count of the number times I've gone into my own code after some time and silently thanked Earlier Me for comments like
//Outputs have to match previous provider, so yes, converting this number to text here is correct
and
'NB loop starting at 1 is deliberate to skip headers
I used to often include links to Sharepoint documents in comments. Stopped doing that after a company "rationalised" their document storage (euphemism for moved to different provider and did not migrate all documents)
CHill60 wrote:
I've gone into my own code after some time and silently thanked Earlier Me for comments like
Same :laugh: :laugh: :laugh:
Jeremy Falcon
-
Questions Is it Maintained? Do you look for it or just explore the code? What do you do if it is not there? Is the code documenting itself good enough?
I just got a copy of some Fortran code I wrote in 1982. Thankfully, I put in a lot of comments. Sadly, I was missing one crucial comment, but I figured it out in about 10 hours. CQ de W5ALT
Walt Fair, Jr.PhD P. E. Comport Computing Specializing in Technical Engineering Software
-
What is an OF, and more to the point, what other post? Inquiring minds want to know (I looked at the thread, and this is your first post). But, I completely agree with your comment.
Charlie Gilley “They who can give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety.” BF, 1759 Has never been more appropriate.
OF = Old Fogey (or some ruder word that starts with F.)
-
I just got a copy of some Fortran code I wrote in 1982. Thankfully, I put in a lot of comments. Sadly, I was missing one crucial comment, but I figured it out in about 10 hours. CQ de W5ALT
Walt Fair, Jr.PhD P. E. Comport Computing Specializing in Technical Engineering Software
Old guru Tony Hoare remarked after reviewing the various proposals for Fortran-77 extensions (some of them went quite far, to phrase it in a polite manner): "I don't know what programming languages will look like in the year 2000, but I know that they will be named 'Fortran'." When I first saw code written to the Fortran-2003 standard, my immediate reaction was: Hoare was right! So my curious question is: How did your 1982 vintage Fortran code fare with a 2024 vintage Fortran compiler? (Fortran 77 was so much delayed that the '77' is a misnomer, so I wouldn't be surprised if 1982 vintage code is not even by the F77 standard.)
Religious freedom is the freedom to say that two plus two make five.