On 7/2/2017 7:27 AM, Vladimir Panteleev via Digitalmars-d wrote:
On Friday, 30 June 2017 at 12:48:12 UTC, Martin Krejcirik wrote:
DMD, Phobos and Druntime open regressions over time:
http://bid.iline.cz/~mk/tmp/regs.png
Used to be stable, but seems to be getting worse since 2016.
One thing that might have contributed to that is that until a year or
two ago, we weren't really checking whether filed bugs were
regressions. As it turns out, a good deal of the time when someone
runs into a bug, they don't even realize that it's some behaviour that
used to work previously. This is why you will occasionally see
recently-filed bugs that are marked as regressions in very old
versions, likely older than since the submitter started using D.
Knowing whether a bug is a regression is useful because then you can
track down the change that caused it, and it's often much easier to
find the bug in a small diff and fix it. The downside is that it makes
the meaning of the "regression" severity less useful when tracking how
many of those issues actually broke someone's code (that we know
about), which is why the regressing version is prefixed in issue
summaries.
One of the biggest issues is that for a small period of time a few years
ago, releases were actually gated on fixing regressions. That stopped
at some point and the backslide has gotten pretty bad. There was a
period where there was exactly one open regression. It's one of my big
disappointments in the current dev/release cycles. That said, since I
really haven't been participating in active development, I tend to just
bite my tongue and say nothing. Holding to a line of no known
regressions is a critical aspect of incrementally better releases.
