Slashdot Mirror
Code Cleanup Culls LibreOffice Cruft
mikejuk writes with an interesting look at what coders can get around to after a few years of creating a free office suite: dealing with many thousands of lines of deprecated code: "Thanks to the efforts of its volunteer taskforce, over half the unused code in LibreOffice has been removed over the past six months. It's good to see this clean-up operation but it does raise questions about the amount of dead code lurking out there in the wild. The scale of the dead code in LibreOffice is shocking, and it probably isn't because the code base is especially bad. Can you imagine this in any other engineering discipline? Oh yes, we built the bridge but there are a few hundred unnecessary iron girders that we forgot to remove... Oh yes, we implemented the new chip but that area over there is just a few thousand transistors we no longer use... and so on." Well, that last one doesn't sound too surprising at all. Exciting to think that LibreOffice (which has worked well for me over the past several years, including under the OpenOffice.org name) has quite so much room for improvement.
There are probably dozens of extra nails that were just hammered in rather than be removed. There are extraneous pieces of lumber.
And a house that was remodeled? I've seen newspaper used as filler. I've seen layers of roofing, with things buried in between layers.
Frankly I don't know what's inside my walls, and I'm not sure I want to know.
I'm mostly surprised that they're still getting performance improvements. It seems like they've done more over the last year than Sun did during the entire time it owned the project to unbloat it.
Oh yes, we built the bridge but there are a few hundred unnecessary iron girders that we forgot to remove...
Well, look at bridges built in the 1800's compared to the ones today. Would we build a modern bridge today using wrought iron links http://en.wikipedia.org/wiki/Clifton_Suspension_Bridge? Each building made in a certain period in a way represents a degree of refinement compared to its predecessors. Better materials, better methods. Buildings in general cannot be "cleaned up" the way code can, where "cruft" today was yesterday's conservative design.
Read a book about the differences in the construction of the World Trade Centers versus the Empire State Building, for example (the WTC has sibling buildings still around using the same techniques, such as the Aon [nee Amoco] Building in Chicago)...
...subsystems replaced in boards by new designers not confident enough to remove the old design, etc.
It sounds crazy, but I work with a real-life example, a beamforming circuit board that utilizes a certain technique, but has all the legacy components utilizing another technique that was never even implemented!
In that case, it wasn't a matter of confidence, but probably corporate sloth - engineers are expensive, and so they figure that paying the board-house more for the extra components per board would be cheaper than getting an engineer to redesign the board.
The quoted section in the summary asks if we could imagine this in other engineering disciplines. As the rest of the summary points out, it happens all the time in microchips. It also happens a lot in civil engineering, including bridge building. Removing things takes work. Unless there's work to be saved by doing it, or some way to profit from selling what's removed as scrap or it's a safety issue to leave it most engineers won't remove old parts of a structure. Consider underground pipes. How often are they removed when they're replaced? If the new ones are being laid down where the old ones went, they'll be replaced. Otherwise, 90% of the time they'll just leave the old ones there. Same goes for just about everything. Old installations of any kind are full of stuff that no longer serves any purpose. Brackets and supports for heavy equipment that isn't used anymore, old wiring and panels, concrete slabs that something mystery object used to sit on, etc. When was the last time you saw anyone take away some 30 ton piece of equipment then pay more money to have the floor where it used to sit un-reinforced? Now, sometimes they do. Usually it's when the place is being sold and the new owners are re-modelling. Other times the owners do decide to do a major cleanup. That's exactly what's being done here with libreoffice. Makes it no different than any other engineering discipline then.
Incidentally, if it's truly "dead" code, then it shouldn't actually be compiled, so it's not like the bridge engineer left in a bunch of extra girders, it's more like he's keeping addendum 6-c to revision 12b of the plans for section 3 in the same file cabinet as revision 13 rather than shifting it to a storage box and warehousing it.
Human DNA (and just about every other species as well) is full of things like inactive duplicate genes (some with slight alterations), pieces of old retroviruses, and other mutations and replication errors that have been "commented out". Plus a whole lot of sequences which we don't know what they're good for yet.
Well, it was an embedded company. This was the first project of such a large scale, and they lacked experience. The manager had worked there for 30 years, and had an electrical engineering background. So it wasn't an ideal situation: he was certainly competent and he could definitely write code, but lacked the experience in software engineering, like how to keep a large software base maintainable. So for example, he had this obsession with not changing code once it was tested. Since all these modules were tested, he was very nervous about changing it. What he failed to see, though, was that more modules were going to be added, and without a clean definition of how this data was to be represented, constant conversions were going to be needed (plus some other things I'm not going into now). Also, automated testing was not a practice there. This is one of the things I worked on introducing at the company, although frankly it was much to late to add it to the existing project. So he was never going to allow to merge the changes back in, because it "broke" tested code.
And I'm not saying I'm all for changing tested code - not at all , but in some circumstances, spending some effort up front can save you a lot of time later on. And I'm sure it would have.
Also, I'm quite confident of my skills. Sure I can still improve, but I surely developed a reputation of writing code that "just works". After 4 years, only a small fraction of the bugs were assigned to me, mostly they were located in parts I didn't write. Mastering the programming language is important, but there are lots of other very important things that matter. You just need a lot of discipline, checking all input, immediately failing when something goes wrong (not letting bad data trickle down the code), clean separations, higher order programming, trying to minimise the interface between modules etc... The list is so large, it becomes more of an intuition.
I hope this clarifies it a bit, I surely wasn't expecting the Spanish Inquisition.
"It's too bad that stupidity isn't painful." - Anton LaVey
Just to be Snarky, I'll point out that the Glasgow Haskell Compiler politely informs me whenever it finds a dead function. Functional languages are light years ahead of anything else when it comes to the Compiler actually being able to reason about the code it's compiling.