Hard to understand 'eval'

[TheSaint]

Hi,

It seems to be strange that give me syntax error inside an eval statement.
I'm looking at it carefully but I can't see any flaw.

Here it's part of the code:

for nn in stn_items:
value= eval('cp.%s' %nn)
if value and (nn in 'log, trash, multithread, verbose, download'):
cfl[wchkey][nn]= chkbool(value)
continue
if value:
cnfg= 'cfl[wchkey][nn]= _%s(value)' %nn
eval(cnfg)

And the output on pdb:

(Pdb) p cnfg
'cfl[wchkey][nn]=_append(value)'
(Pdb) p cfl[wchkey][nn]
False
(Pdb) eval('cfl[wchkey][nn]= _append(value)')
*** SyntaxError: invalid syntax (<string>, line 1)
(Pdb) p value
'230k'
(Pdb) p nn
'append'

Obviously I've an _append() function to convert into decimal the given value.

Other "eval" before this not issuing problems and also rather complicated,
but I'm not seeing the error here.
I'd like to study a class that might get a string and convert it into
function, once it's found inside the program.

 

[bruno.desthuilliers]

Hi,

It seems to be strange that give me syntax error inside an eval statement.
I'm looking at it carefully but I can't see any flaw.

Here it's part of the code:

for nn in stn_items:
value= eval('cp.%s' %nn)

Hem... Not obvious from this snippet, but what's wrong with
getattr(cp, nn) ?

 

[TheSaint]

what's wrong with getattr(cp, nn) ?

The learning curve to get into these programming ways.
Does gettattr run the snippet passed in?
Considering that nn is a name of function, which will be called and (cfl,
value) are the parameters to passed to that function.

I'll spend some bit on getattr use.

 

[TheSaint]

I already see a syntax error when viewing that in Agent... A missing
indent level under the "for"

The program don't complain wrong indentation, I mostly sure a wrong
copy-paste error.
Error doesn't come up there.

. You also don't need the continue if you change the second if into elif.

My mistake, I thought that was improving the loop.

is it an if....elif....elif probing only the first matching case and drop the
remaining checks?

And what type of structure is "cfl"?

You got close, that's a dictionary of dictionaries and I'm trying to updating
it.

wonder what this mysterious _append() function is supposed to be doing;

Append() is a conventional name regarding a file logging.
There would be an option to set a quota of bytes size.

Huh... I presume you mean to convert from a text decimal

it isn't so, the function look at the string to see if ending by K or M,
which means Kbytes or Mbytes. It'll return the decimal conversion.
If the value is set as boolean value, then it will do appending to the log
when it True or stop appending when there's a quota.

def _append(c, v):
RE_BYTE= re.compile(r'^[\d]+(k|m)?$',re.I)
# any number of digit followed by 0 or 1 (k or m), case insensitive
chkbool(v)
if isinstance(v,bool):
c['append']= v
return c
if RE_BYTE.match(value):
k= 1024; M= k * k; v= int(value[:-1])
if value[-1:] == 'k': v= v * k
if value[-1:] == 'm': v= v * m
c['append']= v
return c

All the code could be download at my web site
But this here it's a bit new concept.

 

[Calvin Spealman]

The point here is that eval() use is general frowned upon. If you
don't understand it or the alternatives, then you probably don't
understand it well enough to make the call on using it or not.

If you need just look up an attribute where the name of the attribute
is in a variable, use getattr(obj, attribute_name). If you need to
call a method somewhere, you should have both the name of the method
and the list of arguments to call it with, such as getattr(obj,
methname)(a, b, c). Does this make sense?

 

[John Machin]

Unless the compile/interpret pass is very unoptimized, once a branch
has been taken, the others should be totally skipped.

Don't you mean "Unless the "compile/interpret pass is VERY VERY
BROKEN", as in "omits to drop in a jump to the end of the 'if'
statement after each chunk of action code"?

 

[TheSaint]

such as getattr(obj,
methname)(a, b, c). Does this make sense?

This is big enlightenment Thank you!

I found problem with eval() when it comes to pass quoted strings.
I circumvent that by encapsulating the strings in variable or tuple.
The principle is to have a name which will refers a function somewhere in the
program and to call that function, plus additional data passed in.

In other word I'd expect something:

function_list= ['add' ,'paint', 'read']
for func in function_list:
func(*data)
I tried getattr, and I saw that result. I only investigate a little, so I
still have a small perplexity.

 

[TheSaint]

I don't do regular expressions... and the comment doesn't help
"digit followed by 0 or 1", when 0/1 ARE digits themselves...

That means either none or one letter, of which k or m are allowed.

c is a mutable object; you don't have to return it; the change is
seen by anything holding a reference to the object.

C is a dictionary, it might be omitted, but I'm still not sure if will lose
its state.

Again, as c is mutable, it doesn't need to be returned.

Apparently

Not so apparent, it's doing that, buddy

v = v.strip().lower()

regexp did a fine check and case insensitive, but I like your idea too.

Biggest flaw, in my mind... You are using ONE identifier to control
TWO meanings... a boolean On/Off control AND an integer size limit
control.

That occupy only small variable and as long as python can accept anything not
zero, false or none for an _if_ condition, that might be allowable, I think.
Then if not zero will mean the log will be *appended* to an existing file and
if the value is something that can be converted in decimal value, then this
will set the quota for the log file, as well.
Here below dbg is the log file and if it isn't None then s a valid file path
should work.

dbg= sttng['log']; mode= 'w' # normally write, it writes new
try:
if sttng['append'] > Path.getsize(dbg): mode= 'a'
except (OSError, TypeError): # found a boolean or dbg is new file
pass

So 5 line of code can do me the job smartly, I think

 

[Gabriel Genellina]

It seems to be strange that give me syntax error inside an eval statement.
I'm looking at it carefully but I can't see any flaw.

Here it's part of the code:

for nn in stn_items:
value= eval('cp.%s' %nn)
if value and (nn in 'log, trash, multithread, verbose, download'):
cfl[wchkey][nn]= chkbool(value)
continue
if value:
cnfg= 'cfl[wchkey][nn]= _%s(value)' %nn
eval(cnfg)

And the output on pdb:

(Pdb) p cnfg
'cfl[wchkey][nn]=_append(value)'
(Pdb) p cfl[wchkey][nn]
False
(Pdb) eval('cfl[wchkey][nn]= _append(value)')
*** SyntaxError: invalid syntax (<string>, line 1)

Others have already remarked that your general approach is bad ("don't use eval!", in short). But none has pointed out the actual error in your code: eval can accept only an *expression*, not a statement. eval("x=1") gives the same SyntaxError.
(`exec` is the way to execute statements, but the same caveats apply, so: don't use exec either, use getattr/setattr instead)

 

[Bruno Desthuilliers]

TheSaint a écrit :

The learning curve to get into these programming ways.
Does gettattr run the snippet passed in?

Nope, it just does what the name implies.

Considering that nn is a name of function, which will be called and (cfl,
value) are the parameters to passed to that function.

Everything in Python's an object (at least anything you can bind to a
name), including functions and methods. Once you have a callable object,
you just have to apply the call operator (parens) to call it. In your
case, that would be:

func = getattr(cc, nn, None)
if callable(func):
result = func(cfl, value)
else:
do_whatever_appropriate_here()

I'll spend some bit on getattr use.

Would be wise IMHO.

 

[Bruno Desthuilliers]

TheSaint a écrit :

such as getattr(obj,
methname)(a, b, c). Does this make sense?

This is big enlightenment Thank you!

I found problem with eval() when it comes to pass quoted strings.
I circumvent that by encapsulating the strings in variable or tuple.
The principle is to have a name which will refers a function somewhere in the
program and to call that function, plus additional data passed in.

In other word I'd expect something:

function_list= ['add' ,'paint', 'read']
for func in function_list:
func(*data)

Can't work - function_list is a list of strings, not a list of
functions. If the functions you intend to call are already bound to
names in the current scope, you don't even need any extra lookup
indirection:

def add(*args):
# code here

from some_module import paint

obj = SomeClass()
read = obj.read

functions = [add, paint, read]
args = [1, 2]
for func in functions:
func(*args)

I tried getattr,

getattr is useful when you only have the name of the
function/method/whatever attribute as a string. And a target object
(hint: modules are objects too) of course - if the name lives either in
the global or local namespace, you can access it by name using the dicts
returned by resp. the globals() and locals() functions.

HTH