jqjester

Or perhaps an author fully understands their own work the same way a programmer knows how users will use their programs.

In many cases, isolated testing does not require test doubles. There are alternatives to mocks: as long as the system under test is not manipulating its dependencies, you can get the state you need from the external resources and pass that into your classes.

They don't have to be unbounded to be destructive to this estimation method, just large and unknowable.

I'd rather have two simple things than one complex thing.

In the codebases I've worked on, a lot of the complexity I see arose from people changing the meaning of existing code so they could 'reuse' it.

That sounds fair.

Again, McConnell isn't disagreeing with Bossavit's claims about the Curtis 1986 citation. He concurs! "I never stated that Curtis 1986 was an 'empirical study.' Curtis 1986 is a broad paper that touches on, among other things, differences in programmer productivity. "

The difference is that McConnell thinks that summarizing four studies "offers support for the 10x claim" and Bossavit does not. That's a matter of analysis & evidentiary standards.

I was originally responding to the challenge "aren't you taking Bossavit's claims on trust?"

I'm differentiating between Bossavit's claims about the citations from Bossavit's analysis and arguments. I have to take his claims about the articles and books on trust, because I haven't read them. On the other hand, I don't have to trust his analysis and arguments - I can evaluate those on my own.

In the case of DeMarco and Lister, for example, Bossavit writes "the only 'studies' reported on therein are the programming contests organized by the authors, which took place under loosely controlled conditions (participants were to tackle the exercises at their workplace and concurrently with their work as professional programmers)"

That's the only claim that I have to trust here. Everything else is analysis. McConnell agrees with that description of the citation, but disagrees that it invalidates DeMarco and Lister's research. He thinks that a programming contest that took place under loosely controlled conditions is a legitimate research study.

Sure - I'm pointing to the section titled "A Deeper Dive into the Research Supporting '10x'".

I'm accepting one of Bossavit's claims on trust: he says that of McConnell's eight citations in support of the 10x claim, only two were original research based on empirical studies under controlled conditions (and of those two one only concerned debugging).

I'm willing to trust that because it looks like McConnell mostly agrees. For example, regarding the 1986 Curtis paper, McConnell writes "Bossavit states that, “the 1986 Curtis article does not report on an empirical study.” I never stated that Curtis 1986 was an 'empirical study.'" Regarding DeMarco and Lister, he says "Editorial insinuations aside, that is a correct description of what DeMarco and Lister reported, both in the paper I cited and in Peopleware. Their 1985 study had some of the methodological limitations Sheil’s discussed in 1981".

They disagree about what counts as evidence supporting the 10x claim, but apart from the Card 1987 citation (about which Bassavit was wrong), McConnell confirms Bossavit's claims about every citation.

Steve McConnell himself doesn't dispute Bossavit's claims about the earlier studies (for the most part), he just disagrees about the significance of those claims. See McConnell's rebuttal in which he acknowledges the methodological weaknesses of the citations: http://www.construx.com/10x_Software_Development/Origins_of_... .