csci8980-f21/papers/scp/response-to-reviewers.txt
2020-03-04 19:16:18 -03:00

437 lines
19 KiB
Text

[We thank the reviewers for their comments. Responses interleaved in
brackets. The revision includes a few other improvements, in
particular consistently using standard terms "extrinsic"/"intrinsic"
in place of "raw"/"inherent".]
Ms. Ref. No.: SCICO-D-19-00172 Title: Programming Language Foundations
in Agda Science of Computer Programming
Dear Professor Philip Wadler,
Thank you very much for having submitted to the Science of Computer
Programming Journal.
We have now received reviewers' reports and a recommendation by the
handling editor; based upon the recommendation and the reviews, we
would like to ask you prepare a minor revision taking the comments of
the reviewers into account.
The due date for revision is Feb 14, 2020. If this deadline is out of
sync with other due dates communicated to you outside EES, then please
contact the handling editor to settle this.
If you decide to revise the work, please submit a list of changes or a
rebuttal against each point which is being raised when you submit the
revised manuscript.
To submit a revision, please go to https://ees.elsevier.com/scico/ and
login as an Author. Your username is: wadler@inf.ed.ac.uk If you need
to retrieve password details, please go to:
http://ees.elsevier.com/scico/automail_query.asp
On your Main Menu page is a folder entitled "Submissions Needing
Revision". You will find your submission record there.
Once again, thank you very much for having given us the opportunity to
consider your manuscript.
Mohammad Reza Mousavi, Ph.D. Editor-in-Chief Science of Computer
Programming
Reviewers' comments:
The reviewers have been unanimous in accepting the article for
publication. There are a few remarks to be considered by the authors
for the final version, as pointed out in the reviews.
Reviewer #1: This article reports on experience teaching basic results
in lambda calculus using a proof assistant based on type theory. The
first finding is that the proof style supported by Agda,
i.e. dependently typed programming, need not be more verbose than the
tactic-based style supported by Coq. The second finding is a
functional pearl: constructive proof of progress and preservation
directly yield an interpreter. The third finding is that
formalization of typed lambda calculus based on inherently typed terms
is significantly more succinct than formalization in terms of a
separate typing judgment. Finally, the authors report on flawless
student performance on an exam using the proof assistant.
The last point may not seem surprising to readers with experience
teaching this material as presented in the book Software Foundations
(SF) by Pierce et al, specifically the volume title Programming
Language Foundations. (Or for that matter, teaching related topics
using a proof assistant, e.g., based on the book Concrete Semantics by
Nipkow and Klein.) The authors have written a book, PLFA, drawing on
corresponding chapters of the book by Pierce et al which develops a
more extensive range of programming language theory using Coq.
Section 2 of the article describes PLFA with reference to SF, and
Section 7 describes experience with open source publishing.
At first glance I was concerned that this article might be little more
than a promotion for the book. But in fact the article offers
insightful comparisons between tactical proving and dependently typed
programming style, based on extensive classroom teaching experience.
The book SF introduces quite a few proof tactics and it was not at all
obvious that the same material could be covered perspicuously and
succinctly in Agda. The examples and discussion in Section 3 make a
convincing case for the pedagogical effectiveness of Agda for the
toptics covered. The article acknowledges that this does not
contradict the effectiveness of tactic-based proof assistants for
larger scale developments.
As for the second finding, the authors acknowledge that the idea seems
obvious and can to some extent be discerned in prior publications and
online documents. But it does not seem to have been explicitly
presented in the research literature or textbooks, and it clearly
should be.
The third finding addresses alternatives dubbed extrinsic/intrinsic
typing by Reynolds, which to my knowledge have been viewed as each
having advantages and disadvantages -- thus tradeoffs to be considered
afresh in each situation. The authors make a convincing case for
intrinsic typing by comparing well informed formalizations in both
styles and observing that intrinsic is in some sense more expressive
and also significantly more succinct.
The article is very well written and easy to read. While the
experience report style is somewhat unusual for this particular
journal, there is substantial scientific content and the writing style
is effective in conveying that content. Indeed, it conveys the
authors' enthusiasm for Agda and my review was delayed because I was
keen to get experience with PLFA. A previous version of the paper
appeared in a conference; this article adds substantial material on
additional chapters of the book and on teaching experiences.
Judgment
This is an excellent article worthy of publication in SCP. Below I
mention minor points that could be addressed, but significant revision
is not necessary as the article is already well polished.
Further thoughts
Apropos the volume Verified Functional Algorithms of SF, it makes
heavy use of tactical automation to handle large case analyses
(hundreds of cases, for red-black trees). It would be quite
interesting to see whether at least the simpler algorithms can be
handled well in Agda.
[NOT DONE. Interesting, but we don't have time to do this work.]
page 8 lists logical relations as one of the topics not currently
included in PLFA, but on page 8 under Adequacy a logical relation is
used.
[Clarified by changing "logical relations" to "logical relations
for parametricity"]
page 11 mentions not finding a proof of determinacy of reduction. The
chapter on Normalization of SF, by Tolmach (Norm.v) does so.
[Thanks! Text updated.]
page 12 near the bottom of the page it explains some notations that
occur on previous pages
[Not sure what change is requested here.]
Reviewer #2: Title: Programming Language Foundations in Agda.
Review:
This paper has an interesting structure: it focus on the overall
structure of the book "Programming Language Foundations in Agda", the
choices made and the advantages and difficulties of using Agda to
construct this material (with respect to other tools, in particular
Coq). The technical details are mostly about choices and style of
formalization, and initial results of the application of the material
in a few courses. Let me start by saying that the article is very
pleasant to read: the style is rather informal, but very clear and
provides the reader a good idea of the mindset of the authors that
lead to the choices they made, and also the influences they had along
the way.
A little context regarding my backgroung, to contextualize my
comments: I have been using Coq for roughly three years. During this
time, I have made several exercises of formalization, including the
untyped lambda-calculus (with explicit variable names and with
deBruijn indices), including the Church-Rosser theorem (based on
parallel reductions) and normalisation of left-most reductions. My
experience with Agda is very brief.
In general, I liked the article very much, and I would agree with the
majority of the claims made. The appointed topics below are mostly
comments, suggestions and some comments on a few claims that I found
slightly exaggerated.
Section 2 ---------
There is not much to say about this section, since it is an overview
of the topics that are introduced and proved in each section of the
book. In general, I liked very much the choice of topics and their
sequence.
Page 4, paragraph "Part I, Logical Foundations": suggestion: (briefly
explain monus: the operation is known, but the name 'monus' may not be
as common)
[Not done. It is a standard term, easily looked up. Explaining
it would expand the text more than warranted.]
Page 7, "Discussion": it is mentioned that SF starts with STLC, while
PLFA starts with PCF. It was not discussed the reason to skip a more
restricte, normalizing language (STLC) rather than a more powerful
language (PCF). Could you comment on the motivations for this choice?
It would be very nice to be able to speak about termination proofs and
logical relations at some place (as mentioned in future work).
[Added the requested comment.]
Page 8, "Discussion": It seems that the author claim's that the
substitution in PCF is weaker because it is only applied to closed
terms. However, this is a choice, since it is possible to define
substitution for open terms (only it is usually not worth the effort,
since it is more complex and involves alpha-renaming).
[The choice with substitution of open terms is described later
in the paper, pointing out that the code was too long to include
in the textbook.]
Page 8, "We'd" => "We would" (suggestion: the contraction felt way too
informal...)
[Fixed.]
Section 3 ---------
Regarding the comparison between Agda and Coq, some remarks:
I believe it is clear that Agda's syntax is nicer than Coq's syntax,
in particular very familiar to Haskell programmers. I found it very
interesting the usage of "--" coments to replicate the shape of rules,
mixfix notation and better Unicode support. However, I although I
believe that it was not clear to me that there is a huge advantage of
Agda over Coq in the following two aspects:
-> notation: Coq's notation system, although not as flexible as Agda,
is quite similar in terms of effectiveness: duplication of names
seems like a minor inconvenience, and would not affect the
formalization effort in such a negative way, as claimed by the
authors.
-> style of proof: it seems to me that proving theorems in Agda feels
like programming more than proving in Coq (due to the distinction
between the languages). This does not seem to me as good or bad per
se: it depends on your objectives.
Although it is possible to construct the proof term directly in Coq,
it is not encouraged: the tactic language, although intimidating for
beginners, feels very comprehensive and powerful. Although I agree
that Coq requires an interactive process to be understood, I was not
convinced (by the arguments given in the paper and looking at some
more complex proofs in PLFA) that they are more readable than Coq's:
it seems more a question of preference of style, and desire to use the
same mechanics for programming and proving. Moreover, Coq seems to
"hide" a little better than Adga questions regarding the definition of
equality (you coud hide this at the beginning, and introduce it
later). I agree with the authors that Agda exposes more directly
Propositions as Types and avoid some duplications of Coq.
[The reviewer doesn't ask for specific changes.]
Section 4 ---------
Nice idea. (no problems here)
Section 5 ---------
Page 16, line 45 : "and and" => "and"
[Fixed.]
Page 19, line 34 : it seems that the comparison between intrinsically
and extrinsically-typed lambda terms is not completely fair, in the
sense that there are two main modifications involved: => explicit
names vs deBruijn indices => intrisicly typed vs extrinsicly typed
terms The authors comment that there are actually four possibilities
and comment on some difficulties with the other options. Did you
consider introducing and contrasting each dimension at the time,
rather than making two choices at the same time (in the book)?
[Rephrased to clarify that we think only the two options
we presented work well. The following sentences explain why
the other options are problematic.]
Section 6 ---------
Page 21, line 21: "Similar results were achieved at Edinburgh over the
previous five years".
The presented data seems to suggest that the distinction between the
two systems seems more a question of style, evidencing that it is
possible to substitute Coq for Agda in this scenario. However, I does
not point towards a clear advantage in either way. Maybe an
interesting experiment would be to expose some introduction to both
systems, and try to measure the effort of doing similar tasks in both
platforms.
[The paper includes a suggestion from Nate Foster on similar lines.]
------------------------
Reviewer #3: The paper proposes an alternative to the textbook
Software Foundations (SF) by Pierce et. al. SF uses Coq, while the
authors propose the usage of Agda. The authors claim that, with Coq,
too much effort is spent on learning tactics for proof derivation
instead of learning programming language theory. In contrast, Agda
makes use of no tactics, but its proofs are not much longer than those
in Coq. As a result, the new textbook Programming Language Foundations
in Agda (PLFA) has been written and applied to 3 courses so far
(Edinburgh, Rio de Janeiro, and Padova). In addition to this
contribution to education, the authors mention two other
contributions: the insight that a constructive proof of preservation
and progress automatically produces a prototype evaluator of the
language; and that an intrinsic type relation requires fewer lines of
code and is more expressive than an extrinsic type relation.
The paper reports the development of a solid contribution to the
education of programming language theory and an alternative to the
Software Foundations textbook. On a more scientific note, the usage of
a constructive proof of preservation and progress to produce a
prototype evaluator is a nice insight not clearly described in the
literature.
As suggestions for improvement, I would like to see few comments on
design decisions taken on PLFA: (i) SF begins with a minimal language,
while PLFA begins with a computationally complete language. Why is
that? Apparently, I would regard a minimal language more pedagogical.
(ii) Why lambda calculus instead of Hoare Logic? Was that decision
solely based on the proof automation limitations of Agda or is there a
pedagogical advantage on doing that?
The text is very well written.
Minor fixes: Page 14, line 41. Introduce a line break over the word
"progress." Page 16, line 45. "and and" → "and" Page 20, line
50. "Question 3 give the students the the" → "Question 3 gives the
students the" Overall, I would like to see the text using only the
British spelling or the American spelling (but not both: e.g. flavour,
behavior).
________________
Data in Brief (optional):
We invite you to convert your supplementary data (or a part of it)
into an additional journal publication in Data in Brief, a
multi-disciplinary open access journal. Data in Brief articles are a
fantastic way to describe supplementary data and associated metadata,
or full raw datasets deposited in an external repository, which are
otherwise unnoticed. A Data in Brief article (which will be reviewed,
formatted, indexed, and given a DOI) will make your data easier to
find, reproduce, and cite.
You can submit to Data in Brief via the Science of Computer
Programming submission system when you upload your revised Science of
Computer Programming manuscript. To do so, complete the template and
follow the co-submission instructions found here:
www.elsevier.com/dib-template. If your Science of Computer Programming
manuscript is accepted, your Data in Brief submission will
automatically be transferred to Data in Brief for editorial review and
publication.
Please note: an open access Article Publication Charge (APC) is
payable by the author or research funder to cover the costs associated
with publication in Data in Brief and ensure your data article is
immediately and permanently free to access by all. For the current APC
see:
www.elsevier.com/journals/data-in-brief/2352-3409/open-access-journal
Please contact the Data in Brief editorial office at
dib-me@elsevier.com or visit the Data in Brief homepage
(www.journals.elsevier.com/data-in-brief/) if you have questions or
need further information.
Note: While submitting the revised manuscript, please double check the
author names provided in the submission so that authorship related
changes are made in the revision stage. If your manuscript is
accepted, any authorship change will involve approval from co-authors
and respective editor handling the submission and this may cause a
significant delay in publishing your manuscript.
METHODSX CO-SUBMISSION (OPTIONAL)
If you have customized (a) research method(s) for the project
presented in your Science of Computer Programming article, you are
invited to submit this part of your work as MethodsX article alongside
your revised research article. MethodsX is an independent journal that
publishes the work you have done to develop research methods to your
specific needs or setting. This is an opportunity to get full credit
for the time and money you may have spent on developing research
methods, and to increase the visibility and impact of your work.
How does it work? 1) Fill in the MethodsX article template:
https://www.elsevier.com/MethodsX-template 2) Place all MethodsX files
(including graphical abstract, figures and other relevant files) into
a .zip file and upload this as a 'Method Details (MethodsX) ' item
alongside your revised Science of Computer Programming
manuscript. Please ensure all of your relevant MethodsX documents are
zipped into a single file. 3) If your Science of Computer Programming
research article is accepted, your MethodsX article will automatically
be transferred to MethodsX, where it will be reviewed and published as
a separate article upon acceptance. MethodsX is a fully Open Access
journal, the publication fee is only 520 US$.
Questions? Please contact the MethodsX team at methodsx@elsevier.com.
Example MethodsX articles can be found here:
http://www.sciencedirect.com/science/journal/22150161 Yours sincerely,
INTERACTIVE PLOT VIEWER (OPTIONAL)
Science of Computer Programming features the Interactive Plot Viewer,
see: http://www.elsevier.com/interactiveplots. Interactive Plots
provide easy access to the data behind plots. To include one with your
article, please prepare a .csv file with your plot data and test it
online at
http://authortools.elsevier.com/interactiveplots/verification before
submission as supplementary material. PLEASE NOTE: Science of
Computer Programming would like to enrich online articles by
displaying interactive figures that help the reader to visualize and
explore your research results. For this purpose, we would like to
invite you to upload figures in the MATLAB .FIG file format as
supplementary material to our online submission system. Elsevier will
generate interactive figures from these files and include them with
the online article on SciVerse ScienceDirect. If you wish, you can
submit .FIG files along with your revised submission.
________________
For further assistance, please visit our customer support site at
http://help.elsevier.com/app/answers/list/p/7923. Here you can search
for solutions on a range of topics, find answers to frequently asked
questions and learn more about EES via interactive tutorials. You will
also find our 24/7 support contact details should you need any further
assistance from one of our customer support representatives.