Re: [GiNaC-list] GiNaC::parser parses unary operators in the wrong order
On 20/02/2023, I wrote:
Hi, all. When parsing expressions with sign in the exponent, GiNaC's parses confuses the priorities somehow. Here are some examples:
3^2+1 is parsed as (3^2)+1, which is correct; 3^+2+1 is parsed as 3^(+2+1), which is wrong; 3^-2+1 is parsed as 3^(-2+1), which is wrong; 3^+2/3 is parsed as 3^(+2/3), which is wrong; etc.
So after some more testing I found that GiNaC doesn't only misparse the exponentiation this way, but multiplication too. Things like "3*+2+1" are parsed wrong too. The handling of the unary operators is broken if they appear after other operators. As a separate problem, parsing nested exponentiation, like "2^3^4", results in GiNaC throwing an exception, saying that "power should have exactly 2 operands", which is false: for example ginsh handles "2^3^4" easily and correctly. So, with some work I was able to fix both of these problems. I'm attaching a proposed solution, with tests included. Would it be possible to commit this? I would additionally like to request a point release after this fix; this parsing problem potentially affects our usage of GiNaC, and we'd rather stay with a release version than build our own release tarballs.
Dear Vitaly, Thanks a lot for looking into the parsing issues. On 6/12/23 17:59, Vitaly Magerya wrote:
On 20/02/2023, I wrote:
Hi, all. When parsing expressions with sign in the exponent, GiNaC's parses confuses the priorities somehow. Here are some examples:
3^2+1 is parsed as (3^2)+1, which is correct; 3^+2+1 is parsed as 3^(+2+1), which is wrong; 3^-2+1 is parsed as 3^(-2+1), which is wrong; 3^+2/3 is parsed as 3^(+2/3), which is wrong; etc.
So after some more testing I found that GiNaC doesn't only misparse the exponentiation this way, but multiplication too. Things like "3*+2+1" are parsed wrong too. The handling of the unary operators is broken if they appear after other operators.
As a separate problem, parsing nested exponentiation, like "2^3^4", results in GiNaC throwing an exception, saying that "power should have exactly 2 operands", which is false: for example ginsh handles "2^3^4" easily and correctly.
So, with some work I was able to fix both of these problems. I'm attaching a proposed solution, with tests included. Would it be possible to commit this?
I reviewed your patch and the test vectors and they look good. They make sense. It is apparent that you cared very much for compatibility with ginsh. I have applied the patch.
I would additionally like to request a point release after this fix; this parsing problem potentially affects our usage of GiNaC, and we'd rather stay with a release version than build our own release tarballs.
I understand that GiNaC's parsing is important for your project. Yes, we can do a release soonish. All my best, -richy. -- Richard B. Kreckel <https://in.terlu.de/~kreckel/>
participants (2)
-
Richard B. Kreckel
-
Vitaly Magerya