On Friday, 29 June 2018 at 07:03:52 UTC, Dmitry Olshansky wrote:
I never ever (I think) did something provocative, something to
finally see:
...<rest text is skipped>
My 5 cents inspired by experimenting with D some years ago.
1. Programming became niche oriented and quite diverse. Writing
new language requires significant manpower.
2. D manpower is not sufficient for finishing language in low
level niche. AFAIK Walter estimated manpower around 2013 to be
equivalent of bus factor of 10. This is not enough to deliver
stable language, it will always be "tasted as raw" comparing with
c++.
3. D strategic mistake is ad-hoc design. Some features are added
or extended and because of complexity result in corner cases
(this is exacerbated because sometimes backward compatibility is
preserved and sometimes not). Fixing corner cases sometimes
produces more questions. As a result language has some mess which
is unlikely to be fixed coherently (c++ is at least a documented
mess).
4a. Limited developers' efforts are consumed by fixes and
internal code optimization rather than important issues.
4b. Dips (related to language design) mostly fail because
proposal authors do not write code and developers are busy.
5. My view of D future. Walter and developers will continue to
improve and develop D but at low pace. Low-level niche will be
dominated by c++ as a common denominator. D and some alternative
languages would compete for different parts of this niche. In
long-term low-level niche will be broken into smaller niches with
languages specializing in them. Being "just low level" would be
wrong as "just language". This would raise questions what D goal
is.
I am from area of economic, financial and scientific calculations
used in decision making. It is dominated by python, c++ and
statistical software. In most cases it does not require absolute
speed (except financial markets rt trading, big data processing
or some hard mathematical problems which are relevant for
researchers in top institutions with supercomputers). It is hard
for me to provide arguments for using D (meaning from
professional area view) because c++ can be used for performance
and D is poor in statistical libraries. Because it is applied
area nobody cares whether exceptions have root class or whether
virtual is default.
