Re: library usage statistics

From: Gavin Lambert (boost_at_[hidden])
Date: 2023-04-30 23:57:28

On 1/05/2023 03:19, Peter Dimov wrote:
> Robert Ramey wrote:
>> The basic question that occurs to me is "is this library worth maintaining and
>> keeping in boost".  If dropping the library would break a lot of users code,
>> then the answer is an easy yes.  If dropping the library would result in breaking
>> no code - the answer is an easy no.
>> Otherwise it's a judgement call we can argue about.  But in order to do this we
>> need to know the answer the question: "How much code would be broken if
>> not in Boost.  Hopefully this sheds light on this.
> 
> There's probably a way to answer the question "how many Debian (Ubuntu,
> Fedora...) packages would be broken if we remove library X from Boost", but
> I don't know what it is.

You could in principle trawl the package indexes and find how many 
packages have a direct dependency on "libboost-xyz" (or whatever the 
local package name for a specific library is), and then recursively work 
upwards by finding packages that depend on that package, etc.

It's harder for the Boost libs that don't produce binary packages -- 
often there will be a dependency to a readme or something even when 
there isn't a binary, but not always.  You *might* be able to narrow it 
down if the source packages declare a dependency on specific Boost 
library source packages rather than a monolithic libboost-dev, though.