Changes commited to PARI GIT during the last two weeks

6- logint(t_REAL, 2, &z); z was uninitialized

+ trivialize logint(a, 2^f)

1	0	CHANGES
35	10	src/basemath/ispower.c
16	18	src/test/32/logint
10	12	src/test/in/logint

fix mat bench for 32-bit arch

2	3	src/test/32/mat
2	2	src/test/in/mat

further typos in nfislocalpower (Manvir Jaswal)

8	5	src/basemath/bnflog.c
6	16	src/test/32/nfislocalpower
20	22	src/test/in/nfislocalpower

nfislocalpower: replace n by |n|

3	2	src/basemath/bnflog.c

nfislocalpower: off-by-1 error

1	1	src/basemath/bnflog.c
2	2	src/test/32/nfislocalpower

5- x==0 incorrect for t_COMPLEX x (defined as |x| == 0) [#2708]

was incorrectly assuming we must have real(x) and imag(x) t_REAL to have
|x| == 0 with real or imaginary part non-zero but norm(0.0 + I/2^1000) = 0.E-77

Also fix cx_approx0

1	0	CHANGES
29	22	src/basemath/gen2.c
5	1	src/test/32/mat
5	0	src/test/in/mat

nfislocalpower(K, pr, non-integral element, n)

also optimize for huge n

18	3	src/basemath/bnflog.c
2	1	src/test/32/nfislocalpower
1	0	src/test/in/nfislocalpower

4- nfislocalpower: error when argument not an algebraic integer [#2705]

multiply by denom^n to reduce to the integral case.

1	0	CHANGES
3	1	src/basemath/bnflog.c
3	1	src/test/32/nfislocalpower
4	0	src/test/in/nfislocalpower

update comment

1	1	src/basemath/alglin1.c

rnfgaloisconj_bound: typo in previous commit

2	2	src/basemath/nffactor.c

rnfgaloisconj_bound: need to count complex places twice [#2707]

2	1	CHANGES
2	1	src/basemath/nffactor.c
3	1	src/test/32/rnf
2	0	src/test/in/rnf

galoisfixedfield: doc: explain action of H on the relative extensiion

20	8	src/functions/number_fields_galois/galoisfixedfield

make gen_parapply_slice_zv gc-safe

1	1	doc/usersch4.tex
4	2	src/basemath/alglin3.c

improve doc

12	9	doc/usersch4.tex

improve doc

9	5	doc/usersch4.tex

ellanQ_zv: fix gc

- remove gc_leaf in ellapQ_zv: inefficient (since the stack is cluttered
  from gen_parapply_slice_zv) and useless (doesn't change memory usage)

- add gc in ellQap_u

8	4	src/basemath/elliptic.c

support empty vector in gen_parapply_*

localbitprec(53); lfun(ellinit([1,2]), 100) -> FPE

1	0	CHANGES
1	1	doc/usersch4.tex
15	9	src/basemath/alglin3.c
2	1	src/test/32/lfun
2	0	src/test/in/lfun

z=ffgen(23^3,'z);hyperellcharpoly(t^3+z*t+4) -> SEGV [#2703]

2	0	CHANGES
1	1	src/basemath/ZX.c
2	1	src/test/32/hyperell
3	0	src/test/in/hyperell

Bump version to 2.20.0

3	54	CHANGES
59	0	CHANGES-2.18
1	1	README
1	1	config/version
1	1	doc/refmacro.tex
1	1	src/test/32/version

trace(Mod(x, Mod(1,2)*(x^2+x+1))) ->0 [#2702] (from Manvir Jaswal)

2	1	CHANGES
1	1	src/basemath/FpX.c
1	1	src/basemath/RgX.c
1	0	src/test/32/pol
1	0	src/test/in/pol

doc typo

1	1	src/functions/number_fields_relative/HEADER

add galoisorder

1	0	doc/refcard-nf.tex

update NEW and COMPAT

3	1	COMPAT
11	1	NEW

fix bnfcertify and c_2 phase ("be honest") in bnfinit

Both incorrectly skipped part of the checks in degree> 2. Removing the
last unramified prime above p is only correct if all primes above p have
been included. Initial report and analysis by Manvir Jaswal.

[libpari] prime_over_p_full

3	0	CHANGES
4	0	doc/usersch6.tex
25	3	src/basemath/buch2.c
1	0	src/headers/paridecl.h

CHANGES: Move 2.18 changes to CHANGES-2.18

0	335	CHANGES
334	0	CHANGES-2.18

SPLIT: rename nbtest* variables to be consistent with nsub

General convention:
- nxxx stands for "number of elements in xxx"
- lxxx stands for lg(xxx)
- avoid the name 'l' for anything that is not an 'lg'.

7	7	src/basemath/buch2.c

SPLIT: handle length of subfactorbase cleanly

No bug: when ++lgsub < minss(8, lg(Vbase) - 1) fails once, then nbtest_lim
is set to LONG_MAX and we never hit that line again. So lgsub could be as
large as lg(Vbase) - 1 at most and there is no out-of-bound access in Vbase.

In fact the intent was to go as far as lg(Vbase) so there was a
(harmless) off-by-1 error. New code is less cryptic

16	13	src/basemath/buch2.c

add example

5	1	doc/usersch2.tex