Segfault in tst/testinstall/syntaxtree.tst

[olexandr-konovalov]

This happens in the master branch when all packages are loaded. See e.g. on Travis here: https://travis-ci.org/gap-infra/gap-docker-master-testsuite/jobs/480427212. Started to happen two days ago.

[olexandr-konovalov]

Incidentally, on the same day Semigroups and Utils tests started to fail too:

• https://travis-ci.org/gap-infra/gap-docker-pkg-tests-master/jobs/480676994
• https://travis-ci.org/gap-infra/gap-docker-pkg-tests-master/jobs/480677009

for the same versions of these packages for which they used to work a day before.

[olexandr-konovalov]

The error happening in Semigroups does not sound right at all:

######## > Diff in:
/home/gap/inst/gap-master/pkg/semigroups-3.0.20/tst/standard/congrms.tst:104
# Input is:
IsSubrelation(cong2, cong);
# Expected output:
Error, Semigroups: IsSubrelation: usage,
congruences must be defined over the same semigroup,
# But found:
Error, Z(3)^7Z(3)^7
########
#I  Errors detected while testing

[ChrisJefferson]

So, here's the bug (I think).

src/syntaxtree.c:174 is: value = MakeObjInt(CONST_ADDR_EXPR(expr) + 1, size);. This is passing a pointer inside a bag to MakeObjInt, which internally allocates a new bag, so this pointer moves and awful things happen.

I think in this case we could replace this with EvalIntExpr(expr), which is how the evaluator normally turns expr ints into ints, but I'd prefer an integer expert (so @fingolfin :) ) to sanity-check :)

[ChrisJefferson]

EDIT: No, that's old code.

[ChrisJefferson]

Updated problem (I think)

GET_VALUE_FROM_CURRENT_BODY isn't wired up properly to the syntaxtree (I think), so getting constants isn't producing the right answers. This is used in evaluating Strings (for example).

[fingolfin]

GET_VALUE_FROM_CURRENT_BODY is fine, the problem is that I introduced a silly GC bug, by keeping a pointer into an object across an allocation. Should be fixed in PR #3191