Hooray, a bad subject line! * nixpkgs is on commit d817ccbf832419abdd82bc617e058b8a196f6e98 * guix is on commit 6915d20f05363f61cef9c4ed7ebda3fe5a19325d * GUIX_PACKAGE_PATH is unset FWIW, this was while running the Nix daemon with my "service", but it is not required for the imports. abyayala$ guix import nix ~/re-src/nixpkgs curl ;;; SSAX warning: Skipping PI: xml trace: stdenv.isArm is deprecated after 18.03 trace: lib.zip is deprecated, use lib.zipAttrsWith instead trace: Warning: `showVal` is deprecated and will be removed in the next release, please use `traceSeqN` trace: `mkStrict' is obsolete; use `mkOverride 0' instead. trace: `lib.nixpkgsVersion` is deprecated, use `lib.version` instead! error: stack overflow (possible infinite recursion) Backtrace: In guix/import/snix.scm: 191:9 19 (loop #<input: #{read pipe}# 14> () #f _) 191:9 18 (loop #<input: #{read pipe}# 14> () #f _) 191:9 17 (loop #<input: #{read pipe}# 14> () #f _) 191:9 16 (loop #<input: #{read pipe}# 14> () #f _) 191:9 15 (loop #<input: #{read pipe}# 14> () #f _) 191:9 14 (loop #<input: #{read pipe}# 14> () #f _) 191:9 13 (loop #<input: #{read pipe}# 14> () #f _) 191:9 12 (loop #<input: #{read pipe}# 14> () #f _) 191:9 11 (loop #<input: #{read pipe}# 14> () #f _) 191:9 10 (loop #<input: #{read pipe}# 14> () #f _) 191:9 9 (loop #<input: #{read pipe}# 14> () #f _) 191:9 8 (loop #<input: #{read pipe}# 14> () #f _) 191:9 7 (loop #<input: #{read pipe}# 14> () #f _) 191:9 6 (loop #<input: #{read pipe}# 14> () #f _) 191:9 5 (loop #<input: #{read pipe}# 14> () #f _) 191:9 4 (loop #<input: #{read pipe}# 14> () #f _) 191:9 3 (loop #<input: #{read pipe}# 14> () #f _) 191:9 2 (loop #<input: #{read pipe}# 14> () #f _) 191:9 1 (loop #<input: #{read pipe}# 14> () #f _) In sxml/upstream/SSAX.scm: 1896:21 0 (_ #<input: #{read pipe}# 14> #f #<procedure 7f8b08622…> …) sxml/upstream/SSAX.scm:1896:21: Throw to key `parser-error' with args `(#<input: #{read pipe}# 14> "EOF while reading a token " "reading char data")'.
This bug is resolved by following the current documentation and executing export NIX_REMOTE=daemon . However, even then the import fails: $ guix import nix nixpkgs/ hello ;;; SSAX warning: Skipping PI: xml trace: warning: `stdenv.isArm` is deprecated after 18.03. Please use `stdenv.isAarch32` instead trace: lib.zip is deprecated, use lib.zipAttrsWith instead trace: Warning: `showVal` is deprecated and will be removed in the next release, please use `traceSeqN` trace: `mkStrict' is obsolete; use `mkOverride 0' instead. trace: `lib.nixpkgsVersion` is deprecated, use `lib.version` instead! And after about 15 minutes the import crashes with a stack overflow. Andreas
Sorry, I did not look at the error message well enough, the bug is indeed still the same as reported. Andreas
So, I just discussed with a Nix expert, Profpatsch. Apparently something changed in nixpkgs a while ago, so that the following code does not work any more: $ export NIX_REMOTE=daemon $ nix-instantiate --eval --strict -A hello nixpkgs/ trace: `types.list` is deprecated; use `types.listOf` instead trace: types.optionSet is deprecated; use types.submodule instead trace: Warning: `showVal` is deprecated and will be removed in the next release, please use `traceSeqN` trace: `lib.nixpkgsVersion` is deprecated, use `lib.version` instead! trace: lib.zip is deprecated, use lib.zipAttrsWith instead trace: `mkStrict' is obsolete; use `mkOverride 0' instead. trace: warning: `stdenv.isArm` is deprecated after 18.03. Please use `stdenv.isAarch32` instead trace: warning: `stdenv.isArm` is deprecated after 18.03. Please use `stdenv.isAarch32` instead trace: warning: `stdenv.isArm` is deprecated after 18.03. Please use `stdenv.isAarch32` instead trace: warning: `stdenv.isArm` is deprecated after 18.03. Please use `stdenv.isAarch32` instead error: overrideDerivation not yet supported for functors (use '--show-trace' to show detailed location information) We should have checked the return value... Omitting the "--strict" flag is not a solution, since then lazy evaluation leads to most interesting fields of the result being empty (the return string value is "<CODE>"). A potential solution is evaluating for different package fields separately, but this would require more or less a complete rewrite of the importer. I wonder if the solution would not be to drop it completely... Interestingly enough, the programmatic nature of Nix makes it much more difficult to import packages, since we cannot simply use the abstract syntax tree of the textual representation of a package. Here are a few possible commands to run: $ nix-instantiate --eval --strict -E "with import $PATH_TO_NIXPKGS {}; singular.version" "4.1.1p2" $ nix-instantiate --eval --strict -E "with import $PATH_TO_NIXPKGS {}; map (drv: drv.name) singular.buildInputs" [ "gmp-6.1.2" "ncurses-6.1-20190112" "readline-6.3p08" "ntl-11.3.2" "flint-2.5.2" "cddlib-0.94j" ] $ nix-instantiate --eval --strict -E "with import $PATH_TO_NIXPKGS {}; map (drv: drv.name) singular.nativeBuildInputs" [ "bison-3.3.2" "perl-5.28.2" "pkg-config-0.29.2" "hook" ] (notice that there is now a difference between build and native build inputs, and even more types) $ nix-instantiate --eval --strict -E "with import $PATH_TO_NIXPKGS {}; singular.src" [error message] $ nix-instantiate --eval --strict -E "with import $PATH_TO_NIXPKGS {}; (singular.override { fetchurl = lib.id; }).src" { sha256 = "07x9kri8vl4galik7lr6pscq3c51n8570pyw64i7gbj0m706f7wf"; url = "http://www.mathematik.uni-kl.de/ftp/pub/Math/Singular/SOURCES/4-1-1/singular-4.1.1p2.tar.gz"; } $ nix-instantiate --eval --strict -E "with import $PATH_TO_NIXPKGS {}; (linbox.override { fetchurl = lib.id; }).src" [error message] $ nix-instantiate --eval --strict -E "with import $PATH_TO_NIXPKGS {}; (linbox.override { fetchFromGitHub = lib.id; }).src" { owner = "linbox-team"; repo = "linbox"; rev = "v1.6.3"; sha256 = "10j6dspbsq7d2l4q3y0c1l1xwmaqqba2fxg59q5bhgk9h5d7q571"; } So for the latter two cases, one would need to try both and catch errors. Andreas
Hello,
Andreas Enge <andreas@enge.fr> skribis:
> Omitting the "--strict" flag is not a solution, since then lazy evaluation
> leads to most interesting fields of the result being empty (the return string
> value is "<CODE>").
Do we know exactly why Nix stack-overflows? Perhaps we’re evaluating
fields we shouldn’t evaluate?
It used to work :-), so something must have changed in the Nix
evaluation rules or in how Nixpkgs attribute sets are written. We
should ask the Nix folks for guidance.
Ludo’.
‘guix import nix’ has now been removed: https://issues.guix.gnu.org/47871 Closing! Ludo’.