TODO

* write_proof: do non-binary output; check: should be montgomery or not?
* write_proof does not actually work; convert to public.json, proof.json,
  verification_key.json; output human readable ver key (see ~/temp/factor/)

* documenteren
* setup.py: detect which depends/ are available, also allow to use exes, even .exe windows...
  also: put in version
  
* give better feedback on how to set bitlength (e.g., kaplan-meier as motivating example)
* fixed points: allow negative 

* when running from prompt: if there has ever been an exception and we do ctrl-d,
  proof is not made... (which may make sense...)
  
* clean up swig stuff
* qaptools: get rid of output "*** block" ...

* Are those opt flags still needed? Should be set in cmakefile anyway?
* Setup py should still copy only the exes, manifest decides what goes into the package!
* Add strip to makefile, upload new tar gz
* Catch exception for qapver separately (for sudoku)
* Groth proof/andere proof systems?
* Om de dict zelf copy on write te maken is iig wel idee. Behoud bak ref en update als nodig. Dict.copy en hascopied set. Dan hoef je bak ook niet bij te houden
* Is there a better code indicator than lineno? Misschien lasti? Is lineno idd niet gedef als vanuit interpreter?
  Line number not enough eg recursive ranges. 
* Win32 release van qaptools, minimale doc, output mooier maken

* Energy-saving
   - has a small effect on performance, but not that much: cf
       examples meilof$ time python3 manysquares.py 100000 0
       examples meilof$ time python3 manysquares.py 100000 100
   - remove it? produces counter-intuitive results...
   - or check why it does not help that much
* dealing with huge values, e.g., compute 2^10000 gives segfault
  time python3 exp.py 2 10000; do lincomb.reduce() to -p/2,p/2; let nobackend rerurn 2^64 as modulus?
* TypeError instead of ValueError?
* remove gtest 
* solve double dependency thing?
    * libsnark ../xbyak, ../ate-pairing laten gebruiken? libqap met cmake?
        http://mingw-w64.org/doku.php/download
* does making a source distribution work?
* windows cross-compiling?
* benchmarking: @guarded_and_unguarded
   - multivariate regression as recursive linear regression, BenchmarkResult
     with __add__, __mul__, __sub__
* libsnark: automatically load key given by hash?
* bounds
   - Var(bound=...) -> check and put bound in verification key file
   - verbose option: per computed variable, print computed bounds
     (__file__:__line: "x" has bound [0,20])
   - need to override base type (for bounds) and probably also backend?
* secret santa:
   - amazon lambda, snarkjs
   - assert hashes unequal to each other
* make inhibit_profiling configurable
* rename / operator because it's too dangerous
* check where .value is used: can we avoid all use outside of runtime?

* jupyter, mybinder.org, https://aws.amazon.com/getting-started/projects/build-serverless-web-app-lambda-apigateway-s3-dynamodb-cognito/